Notebooks · Teams, costs and notifications · 31

Load Trace

Name: SHIPIT Agent
Author: SHIPIT

Open in Colab View on GitHub 33 code cells · 17 with saved output

Its imports and arguments are checked against shipit-agent 3.0. The outputs were saved when it was written, for an earlier release.

In [2]

python

from pathlib import Path
import sys

ROOT = (
    Path.cwd().resolve().parent
    if Path.cwd().name == "notebooks"
    else Path.cwd().resolve()
)
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))


from examples.run_multi_tool_agent import build_llm_from_env

llm = build_llm_from_env("bedrock")

In [5]

python

from shipit_agent import Agent, AgentHooks

MODEL_COSTS = {
    "bedrock/openai.gpt-oss-120b-1:0'": {
        "input": 0.15,
        "output": 0.60,
    },  # per 1M tokens
    "gpt-4o": {"input": 2.50, "output": 10.00},
}

hooks = AgentHooks()
costs = {"total_usd": 0.0}


@hooks.on_after_llm
def track_cost(response):
    model = response.metadata.get("model", "")
    rates = MODEL_COSTS.get(model, {"input": 0, "output": 0})
    prompt_cost = response.usage.get("prompt_tokens", 0) / 1_000_000 * rates["input"]
    completion_cost = (
        response.usage.get("completion_tokens", 0) / 1_000_000 * rates["output"]
    )
    costs["total_usd"] += prompt_cost + completion_cost


agent = Agent.with_builtins(
    llm=llm,
    hooks=hooks,
)
data = agent.run("Do something complex with multiple tool calls")
print(f"Total cost: ${costs['total_usd']:.4f}")

Out

Total cost: $0.0000

In [10]

python

from shipit_agent import AsyncAgentRuntime


async def main():
    runtime = AsyncAgentRuntime(
        llm=llm,
        prompt="You are a helpful assistant.",
    )
    state, response = await runtime.run("What is 2 + 2?")
    print(response.content)

In [11]

python

from shipit_agent import FunctionTool, AgentHooks


def search_web(query: str) -> str:
    return f"Results for: {query}"


def fetch_url(url: str) -> str:
    return f"Content of {url}"


runtime = AsyncAgentRuntime(
    llm=llm,
    prompt="You are a research assistant.",
    tools=[
        FunctionTool.from_callable(search_web),
        FunctionTool.from_callable(fetch_url),
    ],
    parallel_tool_execution=True,  # run tools concurrently
    hooks=AgentHooks(),  # attach hooks
    context_window_tokens=128000,  # enable context management
)

state, response = await runtime.run("Research quantum computing advances")

In [19]

python

state

Out

RuntimeState(messages=[Message(role='system', content='You are a research assistant.\n\nAvailable tools:\n- search_web: Function tool for search_web\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is preferable to a heavier tool or model call\n\nRules:\n- pass only the needed arguments\n- prefer these tools for stable utility operations\n- fetch_url: Function tool for fetch_url\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the … (7,174 more characters)

In [20]

python

for m in state.messages:
    if m.role == "tool":
        print(f"{m.name}: {m.content}")

Out

search_web: Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling
search_web: Results for: IBM quantum roadmap 2025 2026 qubits
fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supremacy-100-qubits-0121
fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supersupremacy-100-qubits-0121

In [22]

python

def inspect_runtime_state(state):
    print("\n=== FINAL OUTPUT ===")
    final_payload = state.events[-1].payload if state.events else {}
    print("content:", repr(final_payload.get("content", "")))
    print("output:", repr(final_payload.get("output", "")))

    print("\n=== TOOL RESULTS ===")
    for r in getattr(state, "tool_results", []):
        print(f"- {r.name}: {r.output}")

    print("\n=== REASONING ===")
    for e in getattr(state, "events", []):
        if getattr(e, "type", None) == "reasoning_completed":
            print("-", e.payload.get("content", ""))

    print("\n=== MESSAGES ===")
    for i, m in enumerate(getattr(state, "messages", [])):
        print(f"\n[{i}] role={m.role}")
        print("content:", repr(m.content))
        print("name:", getattr(m, "name", None))
        print("metadata:", m.metadata)

In [23]

python

def get_best_result(state):
    if state.events:
        final_payload = state.events[-1].payload
        final = final_payload.get("content") or final_payload.get("output")
        if final:
            return final

    reasoning = [
        e.payload.get("content", "")
        for e in state.events
        if e.type == "reasoning_completed" and e.payload.get("content")
    ]
    if reasoning:
        return "\n\n".join(reasoning)

    tool_outputs = [f"{r.name}: {r.output}" for r in state.tool_results]
    return "\n".join(tool_outputs)

In [24]

python

inspect_runtime_state(state)

Out

=== FINAL OUTPUT ===
content: ''
output: ''

=== TOOL RESULTS ===
- search_web: Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling
- search_web: Results for: IBM quantum roadmap 2025 2026 qubits
- fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supremacy-100-qubits-0121
- fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supersupremacy-100-qubits-0121

=== REASONING ===
- The user wants to "Research quantum computing advances". Likely they want a summary of recent advances in quantum computing. We need to produce a research summary. Possibly include latest breakthroughs, hardware (e.g., superconducting qubits, trapped ions, photonic), algorithms, error correction, scaling, commercial developments. Should be up-to-date as of 2026 (knowledge cutoff 2024 but we can search web for latest). Use search_web to find recent news. Let's search.
- We need to gather recent advances up to 2026. Summarize key milestones: quantum supremacy demonstrations, error correction, hardware: superconducting qubits (Google, IBM, Rigetti), trapped ions (IonQ, Honeywell), photonic (PsiQuantum), neutral atoms (Pasqal, QuEra), topological qubits (Microsoft). Also software: quantum algorithms, quantum machine learning, variational algorithms, QAOA improvements, quantum error mitigation, quantum volume benchmarks. Also mention scaling of qubit counts: IBM's roadmap: 433 qubits 2025, 1,121 qubits 2026? Actually IBM announced 2025 2,000 qubits? Need verify… (37 more characters)
- We need research quantum computing advances. Need summarize latest developments up to 2026. Use web results. Let's open some.
- We need to fetch the page.

=== MESSAGES ===

[0] role=system
content: 'You are a research assistant.\n\nAvailable tools:\n- search_web: Function tool for search_web\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is preferable to a heavier tool or model call\n\nRules:\n- pass only the needed arguments\n- prefer these tools for stable utility operations\n- fetch_url: Function tool for fetch_url\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is prefe… (134 more characters)
name: None
metadata: {}

[1] role=user
content: 'Research quantum computing advances'
name: None
metadata: {}

[2] role=assistant
content: ''
name: None
metadata: {'model': 'bedrock/openai.gpt-oss-120b-1:0', 'provider': 'litellm', 'tool_calls': [{'id': 'call_1_1', 'name': 'search_web', 'arguments': {'query': 'latest quantum computing advances 2025 2026 quantum supremacy error correction scaling'}}]}

[3] role=tool
content: 'Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling'
name: search_web
metadata: {'tool_call_id': 'call_1_1'}

[4] role=assistant
content: ''
… 27 more lines

In [25]

python

get_best_result(state)

Out

'The user wants to "Research quantum computing advances". Likely they want a summary of recent advances in quantum computing. We need to produce a research summary. Possibly include latest breakthroughs, hardware (e.g., superconducting qubits, trapped ions, photonic), algorithms, error correction, scaling, commercial developments. Should be up-to-date as of 2026 (knowledge cutoff 2024 but we can search web for latest). Use search_web to find recent news. Let\'s search.\n\nWe need to gather recent advances up to 2026. Summarize key milestones: quantum supremacy demonstrations, error correction,… (675 more characters)

In [26]

python

from shipit_agent import Agent, AgentHooks

hooks = AgentHooks()


@hooks.on_before_llm
def log_llm_call(messages, tools):
    print(f"Calling LLM with {len(messages)} messages, {len(tools)} tools")


@hooks.on_after_llm
def track_tokens(response):
    usage = response.usage
    if usage:
        print(f"Tokens: {usage.get('total_tokens', 0)}")


@hooks.on_before_tool
def log_tool_start(name, arguments):
    print(f"Running {name}...")


@hooks.on_after_tool
def log_tool_end(name, result):
    print(f"{name} returned {len(result.output)} chars")


agent = Agent.with_builtins(
    llm=llm,
    hooks=hooks,
)

result = agent.run("What is the weather in Tokyo?")

Out

Calling LLM with 2 messages, 32 tools
Tokens: 12148
Running web_search...
web_search returned 2102 chars
Calling LLM with 4 messages, 32 tools

open_url: Playwright fetch failed for https://weather.com/weather/hourbyhour/l/a2d49e31992a3679724161c3a9029211ecefeda3e1da24489637a5685ded3fdf: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 12967
Running open_url...
open_url returned 4000 chars
Calling LLM with 6 messages, 32 tools

open_url: Playwright fetch failed for https://wttr.in/Tokyo?format=3: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 13767
Running open_url...
open_url returned 14 chars
Calling LLM with 8 messages, 32 tools

open_url: Playwright fetch failed for https://wttr.in/Tokyo?format=%l:+%c+%t+%w+%h: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 13815
Running open_url...
open_url returned 25 chars
Calling LLM with 10 messages, 32 tools
Tokens: 13901

In [27]

python

hooks = AgentHooks()
hooks.before_llm.append(lambda msgs, tools: print("calling LLM"))

In [28]

python

total_cost = {"tokens": 0}


@hooks.on_after_llm
def accumulate(response):
    total_cost["tokens"] += response.usage.get("total_tokens", 0)


agent.run("Do something complex")
print(f"Total tokens used: {total_cost['tokens']}")

Out

Calling LLM with 2 messages, 32 tools
Tokens: 12394
Running ask_user...
ask_user returned 427 chars
Calling LLM with 4 messages, 32 tools
Tokens: 12474
Total tokens used: 0

In [29]

python

import time

last_call = {"time": 0.0}


@hooks.on_before_llm
def rate_limit(messages, tools):
    elapsed = time.time() - last_call["time"]
    if elapsed < 1.0:
        time.sleep(1.0 - elapsed)
    last_call["time"] = time.time()

In [30]

python

@hooks.on_after_tool
def filter_pii(name, result):
    if "email" in result.output.lower():
        print(f"Warning: {name} output may contain PII")

In [31]

python

BLOCKED_TOOLS = {"code_execution", "workspace_files"}


@hooks.on_before_tool
def block_dangerous_tools(name, arguments):
    if name in BLOCKED_TOOLS:
        raise PermissionError(f"Tool {name} is blocked by policy")

In [3]

python

from shipit_agent import Agent

agent = Agent.with_builtins(
    llm=llm,
    parallel_tool_execution=True,  # run tools concurrently
)

# result = agent.run("Search for Python 3.13 release notes and today's weather in NYC")
for event in agent.stream(
    "Search for Python 3.13 release notes and today's weather in NYC"
):
    if event.type == "tool_called":
        print(f"Tool called: {event.payload.get('arguments', {})}")
    elif event.type == "tool_completed":
        print(f"Tool done: {event.payload.get('output', '')[:200]}")
    elif event.type == "reasoning_completed":
        print(f"Reasoning: {event.payload.get('content', '')[:200]}")
    elif event.type == "run_completed":
        print(f"Final: {event.payload.get('content', '')[:500]}")
    else:
        print(f"[{event.type}] {event.message}")

Out

[run_started] Agent run started
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: The user wants two pieces of info: Python 3.13 release notes and today's weather in NYC.

We need to fetch up-to-date info. Use web_search for Python 3.13 release notes (maybe "Python 3.13 release not
Tool called: {'query': 'Python 3.13 release notes', 'max_results': 5}
Tool done: [1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains the new features in Python 3.13 , compared to 3.12
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need Python 3.13 release notes and today's weather in NYC. We have search results for release notes. Need to fetch the release notes page. Also need weather. Likely use web search for weather NYC t
Tool called: {'max_results': 5, 'query': 'NYC weather today'}
Tool done: [1] Manhattan, New York City, New York - The Weather Channel
Today's and tonight's Manhattan, New York City, New York weather forecast, weather conditions and Doppler radar from The Weather Channel an
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need to retrieve Python 3.13 release notes and today's weather in NYC. We have search results. Need to open likely sources for both. For Python, the official docs "What's New In Python 3.13". Let's
Tool called: {'url': 'https://docs.python.org/3/whatsnew/3.13.html', 'max_chars': 2000}
Tool done: index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한국어
Polish | polski
Brazilian Portuguese | Portuguê
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We have Python 3.13 release notes summary. Need also today's weather in NYC. Let's open a weather site to get current data. Use open_url on a reliable source like weather.com but need specific info. U
Tool called: {'url': 'https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf', 'max_chars': 1500}

open_url: Playwright fetch failed for https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf: HTTP 404 from https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf
open_url: urllib fetch failed for https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf: HTTP Error 404: Not Found

Tool done: Error: could not fetch https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf — playwright failed: HTTP 4
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need today's weather in NYC. Could use another source like weather.gov or AccuWeather or simple API. Let's open a more reliable URL. Use National Weather Service.
Tool called: {'url': 'https://forecast.weather.gov/MapClick.php?textField1=40.71&textField2=-74.01', 'max_chars': 2000}
Tool done: HOME
FORECAST 
PAST WEATHER 
SAFETY 
INFORMATION 
EDUCATION 
NEWS 
SEARCH 
ABOUT 
 
View Location Examples

Your local forecast office is

New York, NY
 
News Headlines
Learn All About Being a COCORAH
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: User asks: "Search for Python 3.13 release notes and today's weather in NYC". Need to provide both pieces of info. We have fetched Python release notes page with summary. Could extract key highlights.
Final: **Python 3.13 – Release Notes (high‑level highlights)**  
Source: *What’s New in Python 3.13* (Python documentation)  

- **Release date:** 7 Oct 2024.  
- **Major language changes**
  * New interactive interpreter (a refreshed REPL) with improved UI.
  * Experimental **free‑threaded mode** (PEP 703) – allows running Python code without the GIL for certain workloads.
  * **Just‑In‑Time (JIT) compiler** (PEP 744) – optional performance boost for hot code paths.
  * `locals()` now has defined sema

In [13]

python

from shipit_agent import Agent, FileTraceStore

agent = Agent(
    llm=llm,
    max_iterations=10,
    trace_store=FileTraceStore(root_dir=".shipit_traces"),
    trace_id="run-2026-04-09-alice",
)

In [7]

python

from IPython.display import Markdown, display


def markdown(text):
    display(Markdown(text))

In [ ]

python

for event in agent.stream(
    "Search for Python 3.13 release notes and today's weather in NYC"
):
    if event.type == "tool_called":
        print("Tool called:")
        markdown(f"**Tool Called:** `{event.payload.get('arguments', {})}`")
    elif event.type == "tool_completed":
        print("Tool done:")
        markdown(
            f"**Tool Output:**\n```\n{event.payload.get('output', '')[:1000]}\n```"
        )
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
        markdown(event.payload.get("content", ""))
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

The user asks: "Search for Python 3.13 release notes and today's weather in NYC". So need to provide two pieces: Python 3.13 release notes, and today's weather in NYC. Should gather up-to-date info, so use web_search for both topics. Then open URLs for details. Provide summary.

We need to do web searches. Let's search for "Python 3.13 release notes".

Tool called:

Tool Called: {'query': 'Python 3.13 release notes', 'max_results': 5}

Tool done:

Tool Output:

[1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains the new features in Python 3.13 , compared to 3.12. Python 3.13 was released on October 7, 2024. For full details, see the changelog.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fdocs.python.org%2F3%2Fwhatsnew%2F3.13.html&rut=1efa42f765ec1a7db1a6eeec0a7b9eab118df2821166b88c237d1441e91fed72

[2] Python 3.13 - Release History, LTS & EOL - VersionLog
Python 3.13 Release Notes Python 3.13.0, released on October 7, 2024, introduces major advancements in performance, concurrency, and developer experience.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fversionlog.com%2Fpython%2F3.13%2F&rut=a344c1429deb5f549336315491ff0bac94071ac066ce29d28af00b6eae1eb4b2

[3] cpython/Doc/whatsnew/3.13.rst at main · python/cpython
Summary -- Release Highlights Python 3.13 is a stable release of the Python programming language, with a mix of changes to the language, th

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need both Python 3.13 release notes and today's weather in NYC. Need to fetch weather info. We'll search for NYC weather today.

Tool called:

Tool Called: {'max_results': 5, 'query': 'New York City weather today'}

Tool done:

Tool Output:

[1] New York, NY Weather Forecast | AccuWeather
New York , NY Weather Forecast, with current conditions, wind, air quality, and what to expect for the next 3 days.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.accuweather.com%2Fen%2Fus%2Fnew%2Dyork%2F10021%2Fweather%2Dforecast%2F349727&rut=d12472aa45963856a03b496d2712338ee0d0e68923cc9e83f8d1cf2e302140e1

[2] 10-day weather forecast for Manhattan, New York City, New York, United ...
Be prepared with the most accurate 10-day forecast for Manhattan, New York City , New York , United States with highs, lows, chance of precipitation from The Weather Channel and Weather .com
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fweather.com%2Fen%2DTO%2Fweather%2Ftenday%2Fl%2FManhattan%2BNew%2BYork%2BCity%2BNew%2BYork%2BUnited%2BStates%3FplaceId%3D4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf&rut=e9b5a9661ca09a5e9522694395ef367875be8734ffaaa5b4b52a6a0abf39e7d2

[3] 7-Day Forecast 40.71N 73.98W - National Weather Service
Hazardous We

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide Python 3.13 release notes and today's weather in NYC.

We have search results. Need to open relevant pages to extract info.

Open Python 3.13 what's new page.

Tool called:

Tool Called: {'url': 'https://docs.python.org/3/whatsnew/3.13.html'}

Tool done:

Tool Output:

index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한국어
Polish | polski
Brazilian Portuguese | Português brasileiro
Romanian | Românește
Turkish | Türkçe
Simplified Chinese | 简体中文
Traditional Chinese | 繁體中文
dev (3.15)
3.14.4
3.13
3.12
3.11
3.10
3.9
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
3.0
2.7
2.6
 3.14.4 Documentation » What’s New in Python » What’s New In Python 3.13
  |
… 22 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide both Python 3.13 release notes summary and today's weather in NYC. We have release notes page fetched. Need to extract key points (maybe summary). Also need current weather; we have search results but not content. Need to fetch actual weather page (maybe from weather.com) but that may be dynamic; we could use a simple API like wttr.in which returns text. Let's try open_url to https://wttr.in/New+York?format=3 which gives concise weather.

Tool called:

Tool Called: {'url': 'https://wttr.in/New+York?format=3'}

open_url: Playwright fetch failed for https://wttr.in/New+York?format=3: Page.goto: net::ERR_ABORTED at https://wttr.in/New+York?format=3
Call log:
  - navigating to "https://wttr.in/New+York?format=3", waiting until "domcontentloaded"

Tool done:

Tool Output:

new york: ☀️ +12°C

[step_started] LLM completion started

Python 3.13 – Release notes (summary)

Release date: 7 Oct 2024 (official Python documentation)
Key highlights:
- New interactive interpreter with colored tracebacks.
- PEP 703: experimental free‑threaded mode (optional GIL‑disable).
- PEP 744: basic JIT compiler (disabled by default).
- locals() now has defined semantics when the returned mapping is mutated.
- Type parameters can now have default values.
- Standard‑library changes: new PythonFinalizationError, deprecations in argparse, base64.z85encode/decode, copy.replace, default dbm.sqlite3 backend, new os timer‑fd helpers, CLI for random, etc.
- Security: ssl.create_default_context() now enforces VERIFY_X509_PARTIAL_CHAIN.

(Full details are in the official “What’s New in Python 3.13” page: https://docs.python.org/3/whatsnew/3.13.html)

Today's weather in New York City (as of the time of the query)

Condition: Sunny ☀️
Temperature: +12 °C

Source: https://wttr.in/New+York?format=3

Let me know if you need more detail on any of these topics!

In [10]

python

for event in agent.stream(
    "Search for Python 3.14.4 release notes and today's weather in Warsaw"
):
    if event.type == "tool_called":
        print("Tool called:")
        markdown(f"**Tool Called:** `{event.payload.get('arguments', {})}`")
    elif event.type == "tool_completed":
        print("Tool done:")
        markdown(
            f"**Tool Output:**\n```\n{event.payload.get('output', '')[:1000]}\n```"
        )
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
        markdown(event.payload.get("content", ""))
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide both: Python 3.14.4 release notes, and today's weather in Warsaw. Need fresh data for weather (today's date). Use web_search then open_url for Python release notes, and web_search/open_url for weather. Possibly use a weather site. Let's search.

Tool called:

Tool Called: {'query': 'Python 3.14.4 release notes', 'max_results': 5}

Tool done:

Tool Output:

[1] Python Release Python 3.14.4 | Python.org
Python 3.14.4 is the fourth maintenance release of 3.14, containing around 337 bugfixes, build improvements and documentation changes since 3.14.3. Major new features of the 3.14 series, compared to 3.13
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.python.org%2Fdownloads%2Frelease%2Fpython%2D3144%2F&rut=4b57433242b8a3a3eaf4cad8dec0ed94bd8ef4d04cf800e1b21d98a53c14c9bb

[2] Changelog — Python 3.14.3 documentation
This avoids issues with upgrading Python while it is running. ( Note that such 'in-place' upgrades are not tested.) The tracker remains compatible with subprocesses that use new protocol (that is, subprocesses using Python 3.13.10, 3.14.1 and 3.15). gh-142214: Fix two regressions in dataclasses in Python 3.14.1 related to annotations.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fdocs.python.org%2F3%2Fwhatsnew%2Fchangelog.html&rut=5053dd15419ac522349e2e7b5efefbd7c2a7ce91a4c44a9a6458aa9a91282d64

[3] Python 3.15.0a8, 3.14.4 and 3.13.

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

The user asks: "Search for Python 3.14.4 release notes and today's weather in Warsaw". Need to provide both pieces of information.

We have search results for Python 3.14.4 release notes. Let's open the first result.

Tool called:

Tool Called: {'url': 'https://www.python.org/downloads/release/python-3144/'}

Tool done:

Tool Output:

Skip to content
Python
PSF
Docs
PyPI
Jobs
Community
Donate
 
Search This Site
 GO
Socialize
About
 
Downloads
 
Documentation
 
Community
 
Success Stories
 
News
 
Events
Python 3.14.4

Release date: April 7, 2026

This is fourth maintenance release of Python 3.14

Python 3.14.4 is the fourth maintenance release of 3.14, containing around 337 bugfixes, build improvements and documentation changes since 3.14.3.

Major new features of the 3.14 series, compared to 3.13

Some of the major new features and changes in Python 3.14 are:

New features
… 7 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

The user wants both Python 3.14.4 release notes and today's weather in Warsaw. We have fetched release notes page; need to summarize key release notes. Also need today's weather in Warsaw. Need to fetch weather info from a source (e.g., weather.com, openweather). Use web_search to find "Warsaw weather today".

Tool called:

Tool Called: {'max_results': 5, 'query': 'Warsaw weather today'}

Tool done:

Tool Output:

[1] Today's Weather in Warszawa - Hourly Forecast and Conditions
Get the latest hourly weather updates for Warszawa today . Detailed forecast including temperature, wind, rain, snow, and UV index. Stay informed about today's weather conditions in Warszawa.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.easeweather.com%2Feurope%2Fpoland%2Fmazovia%2Fwarszawa%2Ftoday&rut=3dedf7325907f92328e940c240a1553e2d7dd1d46f2891d49f75ef4cf9f46ae9

[2] Warsaw Weather Forecast
12 day Warsaw Weather Forecast. Live Weather Warnings, hourly weather updates. Accurate Warsaw weather today , forecast for sun, rain, wind and temperature.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.weather%2Dforecast.com%2Flocations%2FWarsaw%2Fforecasts%2Flatest&rut=f512cd7bae35b6d3a813f459fa24d6236565d45c9fdf79fb87b30da892ae2d30

[3] Warsaw, Masovia, Poland Hourly Weather | AccuWeather
Hourly weather forecast in Warsaw , Masovia, Poland. Check current conditions in Warsaw , Masovia, Poland with radar, hourly, and more.
U

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide Python 3.14.4 release notes and today's weather in Warsaw. Already have release notes page from python.org. Need to summarize key points. For weather, need current data. Likely need to open one of the weather pages to get actual data. Let's open a reliable source like weather.com or accuweather. Use open_url on one result.

Tool called:

Tool Called: {'url': 'https://weather.com/weather/today/l/Warsaw+Poland?canonicalCityId=7ed0ae4e25e1adb77367c6f70331d838'}

Tool done:

Tool Output:

Home
Today
Hourly
10 Day
Monthly
Radar
Video
Explore
Severe Weather
Hurricane Central
Tornado Central
Safety & Prep
Health & Wellness
Allergy Tracker
Air Quality Index
Cold & Flu Tracker
Skin
Lifestyle
Home & Garden
Pets & Animals
Recipes
Space & Skywatching
Travel & Outdoors
Product Reviews & Deals
Shop
Home
Today
Hourly
10 Day
Monthly
Radar
Video
Explore
Severe Weather
Hurricane Central
Tornado Central
Safety & Prep
Health & Wellness
… 18 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide the Python 3.14.4 release notes and today's weather in Warsaw. We have Python release notes page content; we need to summarize. For weather, the page we opened is weather.com but shows generic data; need current date. The data shows as of 9:00 pm CEST, temperature 48°F, fair. Might be enough. Provide summary.

We should ensure today's date (April 12, 2026). Provide weather details: temperature, conditions, wind, humidity, forecast.

Let's extract relevant info from weather page: In the snippet we see "As of 9:00 pm CEST 48° Fair". Also hourly shows now 46°, clear, etc. Provide summary.

Now craft answer.

Python 3.14.4 Release notes (released April 7 2026)

Overview
Python 3.14.4 is the fourth maintenance release of the 3.14 series. It bundles ≈ 337 bug‑fixes, assorted build improvements and documentation updates over 3.14.3.

Key highlights

Area	What’s new / changed
Free‑threaded interpreter	PEP 779 – free‑threaded Python is now officially supported.
Deferred annotation evaluation	PEP 649 – annotations are evaluated lazily, fixing long‑standing semantics issues.
Template string literals	PEP 750 – “t‑strings” give you a custom‑string‑processing syntax that mirrors f‑strings.
Multiple interpreters	PEP 734 – the stdlib now ships a multiple‑interpreter module.
Zstandard compression	PEP 784 – new `compression.zstd` module for Zstandard algorithm support.
Exception syntax	PEP 758 – `except`/`except*` can omit the surrounding brackets.
Debugger interface	PEP 768 – a zero‑overhead external debugger interface for CPython.
UUID extensions	UUID v6‑v8 added; generation of v3‑v5 now up to 40 % faster.
Return/break/continue in finally	PEP 765 disallows these statements from exiting a `finally` block.
C‑API improvements	PEP 741 – revamped C‑API for configuring Python.
New interpreter type	An optional, high‑performance interpreter for newer compilers (opt‑in).
Error‑message upgrades	More informative tracebacks and diagnostics.
Built‑in HMAC	Formal‑verification via the HACL* project.
Async process inspector	New CLI to inspect running Python processes using asynchronous tasks.
Remote pdb	`pdb` can now attach remotely to a live process.
Build changes	• No longer provides PGP signatures – Sigstore is recommended (PEP 761). • macOS/Windows binaries now include an experimental JIT. • Official Android binary releases are now available.
Installer shift	The traditional Windows installer is being replaced by a new Python Install Manager (available via Windows Store or direct download).

For a full, line‑by‑line changelog see the “Full Changelog” section on the release page or the “What’s new in Python 3.14” docs.

Warsaw (Poland) weather – 12 April 2026

Parameter	Value (as of the latest observation ≈ 21:00 CEST)
Temperature	48 °F (≈ 9 °C) – Fair conditions
Feels like	45 °F (≈ 7 °C)
Current condition	Clear skies
Wind	7 mph (≈ 11 km/h) from ESE
Humidity	45 %
… 16 more lines

In [17]

python

from shipit_agent.tools import WebSearchTool, OpenURLTool

agent = Agent(
    llm=llm,
    tools=[WebSearchTool(), OpenURLTool()],
    max_iterations=10,
    trace_store=FileTraceStore(root_dir=".shipit_traces"),
    trace_id="run-2026-04-09-alice",
)

In [16]

python

for event in agent.stream("Search for new about shipit agent and what is shipiit ai"):
    if event.type == "tool_called":
        print("Tool called:")
    elif event.type == "tool_completed":
        print("Tool done:")
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

ShipIt AI – Quick‑look (what it is)

Feature	Details
Core idea	An “AI‑developer‑workspace” that turns a plain‑English request into a full software‑delivery pipeline: parse → implement → review → fix → test → PR → ship.
AI engines	Supports up to four different generators (Claude, OpenAI Codex, Google Gemini, and a custom “ShipIt CLI” that can route to Bedrock, Ollama, Groq, Together, OpenRouter, etc.) and > 100 models.
Repo‑aware editing	The agent reads the entire repository, validates paths, and prevents out‑of‑scope edits.
Visible delivery loop	Every step (code generation, auto‑review, test run, commit) is shown in a UI/graph so the developer can see what the agent is doing, not just the final output.
Multi‑messenger integration	The UI can spin up Slack, Discord, and Telegram bots that listen for commands (e.g., “@shipit run the test suite”) and reply with live status updates.
Built‑in security / QA	Runs static analysis (Bandit, Semgrep), secret detection, license compliance, container scans, and an AI‑driven penetration‑testing suite (OWASP 2025 coverage).
Extensible workflow	30+ step‑type templates, drag‑and‑drop visual pipeline builder, 17+ MCP integrations (GitHub, GitLab, Jira, DBs, Playwright, etc.).
Public‑facing pages	• Product landing: https://shipiit.com/ • LinkedIn announcement (4 days ago): https://www.linkedin.com/posts/shipiit_shipit-a-powerful-ai-coding-agent-with-activity-7447720216366682112-1Z6I • GitHub “ShipIt‑AI” graph project: https://github.com/ship-it-ops/ShipIt-AI • “ShipIt Clean” code‑analysis spinoff: https://shipitclean.com/ • Devpost hackathon entry (2018): https://devpost.com/software/shipit

Recent news & activity (the “new about ShipIt agent”)

Date (≈)	Source	Highlight
4 days ago	LinkedIn post by ShipIt (35 followers)	Announces a “powerful AI coding agent with a beautiful modern UI” that works across Slack, Discord and Telegram. Shows a concrete workflow where a user asks the bot to run a test suite and receives a live‑streamed result. Includes a wait‑list link to early‑access the product.
2024‑03 (approx.)	ShipIt product site (shipiit.com)	Updated landing page now shows four AI engines, 100+ models, 30+ visual workflow steps and 17+ integrations. Emphasises “repo‑aware before editing”, “visible delivery graph”, and “operator‑first product”.
2024‑02	ShipIt‑Clean website launch	Introduces a separate service that runs multiple independent AI agents to audit a codebase (feature completeness, architecture, performance, developer experience, scalability) and output a prioritized Opportunity Report.
2024‑02‑20 (GitHub commit)	GitHub – ship‑it‑ops/ShipIt‑AI	Open‑source project that maps a software ecosystem into a Neo4j knowledge graph. Connectors pull data from GitHub, Kubernetes, etc., exposing a queryable graph for downstream AI agents (Claude, Gemini, etc.).
2023‑12 (Devpost)	Devpost “ShipIt” hackathon project	Earlier incarnation of “ShipIt” that combined a smart contract on Ethereum with a ML model predicting project completion time. Demonstrates the team’s long‑standing interest in AI‑assisted project delivery.

No major press releases have been published in mainstream tech news outlets yet; the most visible public updates are the LinkedIn announcement and the product‑site refresh that rolled out in early 2024.

TL;DR Summary

ShipIt AI is an end‑to‑end AI‑powered development platform that turns natural‑language prompts into a full CI/CD‑style workflow, complete with code generation, automated review, testing, security scanning, and PR creation.
It runs multiple large‑model back‑ends (Claude, Codex, Gemini, custom CLI) and is tightly integrated with your code repository, so … output trimmed

Load Trace

Here's how the tracing works and how to view traces nicely like LangSmith/LangGraph

In [18]

python

store = FileTraceStore(root_dir=".shipit_traces")
trace = store.load("run-2026-04-09-alice")

# Pretty print like LangGraph/LangSmith
for i, event in enumerate(trace.events):
    indent = (
        "  "
        if event.type in ("tool_called", "tool_completed", "reasoning_completed")
        else ""
    )

    if event.type == "run_started":
        print(f"🟢 Run Started: {event.payload.get('prompt', '')}")
    elif event.type == "step_started":
        print(f"\n📍 Step {event.payload.get('iteration', '?')}")
    elif event.type == "reasoning_completed":
        print(f"  💭 Reasoning: {event.payload.get('content', '')[:150]}...")
    elif event.type == "tool_called":
        print(f"  🔧 → {event.payload.get('arguments', {})}")
    elif event.type == "tool_completed":
        print(f"  ✅ ← {event.payload.get('output', '')[:150]}")
    elif event.type == "tool_failed":
        print(f"  ❌ Failed: {event.payload.get('error', '')}")
    elif event.type == "run_completed":
        usage = event.payload.get("usage", {})
        print(f"\n🏁 Done — tokens: {usage.get('total_tokens', 0)}")
        print(f"   Output: {event.payload.get('content', '')[:300]}")
    else:
        print(f"  [{event.type}] {event.message}")

Out

🟢 Run Started: Search for Python 3.13 release notes and today's weather in NYC

📍 Step 1
  [reasoning_started] Model reasoning started
  💭 Reasoning: The user asks: "Search for Python 3.13 release notes and today's weather in NYC". So need to provide two pieces: Python 3.13 release notes, and today'...
  🔧 → {'query': 'Python 3.13 release notes', 'max_results': 5}
  ✅ ← [1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains 

📍 Step 2
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need both Python 3.13 release notes and today's weather in NYC. Need to fetch weather info. We'll search for NYC weather today....
  🔧 → {'max_results': 5, 'query': 'New York City weather today'}
  ✅ ← [1] New York, NY Weather Forecast | AccuWeather
New York , NY Weather Forecast, with current conditions, wind, air quality, and what to expect for the

📍 Step 3
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need to provide Python 3.13 release notes and today's weather in NYC.

We have search results. Need to open relevant pages to extract info.

Open P...
  🔧 → {'url': 'https://docs.python.org/3/whatsnew/3.13.html'}
  ✅ ← index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한

📍 Step 4
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need to provide both Python 3.13 release notes summary and today's weather in NYC. We have release notes page fetched. Need to extract key points (...
… 420 more lines

In [19]

python

from shipit_agent import Agent

# ToolSearchTool is included in Agent.with_builtins automatically
agent = Agent.with_builtins(llm=llm)

result = agent.run(
    "Use tool_search to find the right tool for fetching a specific URL, "
    "then call that tool on https://example.com"
)

In [21]

python

markdown(f"**Final Result:**\n```\n{result.output}\n```")

Out

Final Result:

Here’s the content fetched from **https://example.com** (truncated to the first 1 000 characters):

Example Domain

This domain is for use in documentation examples without needing permission. Avoid use in operations.

Learn more

In [4]

python

# ToolSearchTool is included in Agent.with_builtins automatically
from shipit_agent import Agent

agent = Agent.with_builtins(llm=llm)

In [5]

python

from pydantic import BaseModel
from shipit_agent import Agent


class MovieReview(BaseModel):
    title: str
    rating: float
    summary: str
    earning: int
    pros: list[str]
    cons: list[str]
    casts: list[str]


result = agent.run("Review The Dhurandhar: The Revenge", output_schema=MovieReview)

review = result.parsed  # MovieReview instance
review.title  # "The Dhurandhar: The Revenge"
review.rating  # 9.5
review.earning  # 1000000
review.pros  # ["Groundbreaking visuals", ...]
review.cons  # ["Weak storyline", ...]
review.casts  # ["Actor 1", "Actor 2", ...]
type(review)  # <class 'MovieReview'>

Out

__main__.MovieReview

In [9]

python

print(f"Title: {review.title}")
print(f"Rating: {review.rating}")
print(f"Earning: ${review.earning}")
print("Pros:")
for p in review.pros:
    print(f"- {p}")
print("Cons:")
for c in review.cons:
    print(f"- {c}")
print("Casts:")
for cast in review.casts:
    print(f"- {cast}")

Out

Title: Dhurandhar: The Revenge
Rating: 3.5
Earning: $16822200000
Pros:
- Strong lead performance by Ranveer Singh
- Impressive supporting cast (Arjun Rampal, Sanjay Dutt, R. Madhavan, Sara Arjun)
- High‑production values and expansive action set‑pieces
- Well‑received soundtrack and technical craftsmanship
- Record‑breaking box‑office performance
Cons:
- Overly long (nearly 4 hours) leading to pacing issues
- Graphic violence and gore that alienated many viewers
- Criticized for overt nationalist messaging and propaganda
- Mixed critical scores (35 % Tomatometer on Rotten Tomatoes)
- Limited appeal outside the core Indian market due to political tone
Casts:
- Ranveer Singh
- Arjun Rampal
- Sanjay Dutt
- R. Madhavan
- Sara Arjun
- Rakesh Bedi
- Gaurav Gera
- Danish Pandor
- Manav Gohil

In [28]

python

resultreview = result.parsed

if review is None:
    print("⚠️ No structured output, generating fallback...")

    review = MovieReview(
        title="The Dhurandhar: The Revenge",
        rating=6.5,
        summary="A fictional action drama centered on revenge with intense sequences.",
        earning=50000000,
        pros=["Action scenes", "Lead performance"],
        cons=["Predictable story", "Weak pacing"],
        casts=["Actor A", "Actor B"],
    )

Out

⚠️ No structured output, generating fallback...

In [24]

python

review = result.parsed

In [25]

python

review

In [26]

python

from pydantic import BaseModel
from shipit_agent import Agent


class MovieReview(BaseModel):
    title: str
    rating: float
    summary: str
    pros: list[str]
    cons: list[str]


agent = Agent.with_builtins(llm=llm)
result = agent.run("Review The Matrix", output_schema=MovieReview)

review = result.parsed  # MovieReview instance
review.title  # "The Matrix"
review.rating  # 9.5
review.pros  # ["Groundbreaking visuals", ...]
type(review)  # <class 'MovieReview'>

Out

__main__.MovieReview

Notebooks · Teams, costs and notifications · 31

Load Trace

Open in Colab View on GitHub 33 code cells · 17 with saved output

Its imports and arguments are checked against shipit-agent 3.0. The outputs were saved when it was written, for an earlier release.

In [2]

python

from pathlib import Path
import sys

ROOT = (
    Path.cwd().resolve().parent
    if Path.cwd().name == "notebooks"
    else Path.cwd().resolve()
)
if str(ROOT) not in sys.path:
    sys.path.insert(0, str(ROOT))


from examples.run_multi_tool_agent import build_llm_from_env

llm = build_llm_from_env("bedrock")

In [5]

python

from shipit_agent import Agent, AgentHooks

MODEL_COSTS = {
    "bedrock/openai.gpt-oss-120b-1:0'": {
        "input": 0.15,
        "output": 0.60,
    },  # per 1M tokens
    "gpt-4o": {"input": 2.50, "output": 10.00},
}

hooks = AgentHooks()
costs = {"total_usd": 0.0}


@hooks.on_after_llm
def track_cost(response):
    model = response.metadata.get("model", "")
    rates = MODEL_COSTS.get(model, {"input": 0, "output": 0})
    prompt_cost = response.usage.get("prompt_tokens", 0) / 1_000_000 * rates["input"]
    completion_cost = (
        response.usage.get("completion_tokens", 0) / 1_000_000 * rates["output"]
    )
    costs["total_usd"] += prompt_cost + completion_cost


agent = Agent.with_builtins(
    llm=llm,
    hooks=hooks,
)
data = agent.run("Do something complex with multiple tool calls")
print(f"Total cost: ${costs['total_usd']:.4f}")

Out

Total cost: $0.0000

In [10]

python

from shipit_agent import AsyncAgentRuntime


async def main():
    runtime = AsyncAgentRuntime(
        llm=llm,
        prompt="You are a helpful assistant.",
    )
    state, response = await runtime.run("What is 2 + 2?")
    print(response.content)

In [11]

python

from shipit_agent import FunctionTool, AgentHooks


def search_web(query: str) -> str:
    return f"Results for: {query}"


def fetch_url(url: str) -> str:
    return f"Content of {url}"


runtime = AsyncAgentRuntime(
    llm=llm,
    prompt="You are a research assistant.",
    tools=[
        FunctionTool.from_callable(search_web),
        FunctionTool.from_callable(fetch_url),
    ],
    parallel_tool_execution=True,  # run tools concurrently
    hooks=AgentHooks(),  # attach hooks
    context_window_tokens=128000,  # enable context management
)

state, response = await runtime.run("Research quantum computing advances")

In [19]

python

state

Out

RuntimeState(messages=[Message(role='system', content='You are a research assistant.\n\nAvailable tools:\n- search_web: Function tool for search_web\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is preferable to a heavier tool or model call\n\nRules:\n- pass only the needed arguments\n- prefer these tools for stable utility operations\n- fetch_url: Function tool for fetch_url\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the … (7,174 more characters)

In [20]

python

for m in state.messages:
    if m.role == "tool":
        print(f"{m.name}: {m.content}")

Out

search_web: Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling
search_web: Results for: IBM quantum roadmap 2025 2026 qubits
fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supremacy-100-qubits-0121
fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supersupremacy-100-qubits-0121

In [22]

python

def inspect_runtime_state(state):
    print("\n=== FINAL OUTPUT ===")
    final_payload = state.events[-1].payload if state.events else {}
    print("content:", repr(final_payload.get("content", "")))
    print("output:", repr(final_payload.get("output", "")))

    print("\n=== TOOL RESULTS ===")
    for r in getattr(state, "tool_results", []):
        print(f"- {r.name}: {r.output}")

    print("\n=== REASONING ===")
    for e in getattr(state, "events", []):
        if getattr(e, "type", None) == "reasoning_completed":
            print("-", e.payload.get("content", ""))

    print("\n=== MESSAGES ===")
    for i, m in enumerate(getattr(state, "messages", [])):
        print(f"\n[{i}] role={m.role}")
        print("content:", repr(m.content))
        print("name:", getattr(m, "name", None))
        print("metadata:", m.metadata)

In [23]

python

def get_best_result(state):
    if state.events:
        final_payload = state.events[-1].payload
        final = final_payload.get("content") or final_payload.get("output")
        if final:
            return final

    reasoning = [
        e.payload.get("content", "")
        for e in state.events
        if e.type == "reasoning_completed" and e.payload.get("content")
    ]
    if reasoning:
        return "\n\n".join(reasoning)

    tool_outputs = [f"{r.name}: {r.output}" for r in state.tool_results]
    return "\n".join(tool_outputs)

In [24]

python

inspect_runtime_state(state)

Out

=== FINAL OUTPUT ===
content: ''
output: ''

=== TOOL RESULTS ===
- search_web: Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling
- search_web: Results for: IBM quantum roadmap 2025 2026 qubits
- fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supremacy-100-qubits-0121
- fetch_url: Content of https://news.mit.edu/2025/quantum-computer-supersupremacy-100-qubits-0121

=== REASONING ===
- The user wants to "Research quantum computing advances". Likely they want a summary of recent advances in quantum computing. We need to produce a research summary. Possibly include latest breakthroughs, hardware (e.g., superconducting qubits, trapped ions, photonic), algorithms, error correction, scaling, commercial developments. Should be up-to-date as of 2026 (knowledge cutoff 2024 but we can search web for latest). Use search_web to find recent news. Let's search.
- We need to gather recent advances up to 2026. Summarize key milestones: quantum supremacy demonstrations, error correction, hardware: superconducting qubits (Google, IBM, Rigetti), trapped ions (IonQ, Honeywell), photonic (PsiQuantum), neutral atoms (Pasqal, QuEra), topological qubits (Microsoft). Also software: quantum algorithms, quantum machine learning, variational algorithms, QAOA improvements, quantum error mitigation, quantum volume benchmarks. Also mention scaling of qubit counts: IBM's roadmap: 433 qubits 2025, 1,121 qubits 2026? Actually IBM announced 2025 2,000 qubits? Need verify… (37 more characters)
- We need research quantum computing advances. Need summarize latest developments up to 2026. Use web results. Let's open some.
- We need to fetch the page.

=== MESSAGES ===

[0] role=system
content: 'You are a research assistant.\n\nAvailable tools:\n- search_web: Function tool for search_web\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is preferable to a heavier tool or model call\n\nRules:\n- pass only the needed arguments\n- prefer these tools for stable utility operations\n- fetch_url: Function tool for fetch_url\n  Guidance: Call a direct Python function tool.\n\nUse this when:\n- a deterministic local function can solve the task directly\n- code-free execution is prefe… (134 more characters)
name: None
metadata: {}

[1] role=user
content: 'Research quantum computing advances'
name: None
metadata: {}

[2] role=assistant
content: ''
name: None
metadata: {'model': 'bedrock/openai.gpt-oss-120b-1:0', 'provider': 'litellm', 'tool_calls': [{'id': 'call_1_1', 'name': 'search_web', 'arguments': {'query': 'latest quantum computing advances 2025 2026 quantum supremacy error correction scaling'}}]}

[3] role=tool
content: 'Results for: latest quantum computing advances 2025 2026 quantum supremacy error correction scaling'
name: search_web
metadata: {'tool_call_id': 'call_1_1'}

[4] role=assistant
content: ''
… 27 more lines

In [25]

python

get_best_result(state)

Out

'The user wants to "Research quantum computing advances". Likely they want a summary of recent advances in quantum computing. We need to produce a research summary. Possibly include latest breakthroughs, hardware (e.g., superconducting qubits, trapped ions, photonic), algorithms, error correction, scaling, commercial developments. Should be up-to-date as of 2026 (knowledge cutoff 2024 but we can search web for latest). Use search_web to find recent news. Let\'s search.\n\nWe need to gather recent advances up to 2026. Summarize key milestones: quantum supremacy demonstrations, error correction,… (675 more characters)

In [26]

python

from shipit_agent import Agent, AgentHooks

hooks = AgentHooks()


@hooks.on_before_llm
def log_llm_call(messages, tools):
    print(f"Calling LLM with {len(messages)} messages, {len(tools)} tools")


@hooks.on_after_llm
def track_tokens(response):
    usage = response.usage
    if usage:
        print(f"Tokens: {usage.get('total_tokens', 0)}")


@hooks.on_before_tool
def log_tool_start(name, arguments):
    print(f"Running {name}...")


@hooks.on_after_tool
def log_tool_end(name, result):
    print(f"{name} returned {len(result.output)} chars")


agent = Agent.with_builtins(
    llm=llm,
    hooks=hooks,
)

result = agent.run("What is the weather in Tokyo?")

Out

Calling LLM with 2 messages, 32 tools
Tokens: 12148
Running web_search...
web_search returned 2102 chars
Calling LLM with 4 messages, 32 tools

open_url: Playwright fetch failed for https://weather.com/weather/hourbyhour/l/a2d49e31992a3679724161c3a9029211ecefeda3e1da24489637a5685ded3fdf: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 12967
Running open_url...
open_url returned 4000 chars
Calling LLM with 6 messages, 32 tools

open_url: Playwright fetch failed for https://wttr.in/Tokyo?format=3: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 13767
Running open_url...
open_url returned 14 chars
Calling LLM with 8 messages, 32 tools

open_url: Playwright fetch failed for https://wttr.in/Tokyo?format=%l:+%c+%t+%w+%h: It looks like you are using Playwright Sync API inside the asyncio loop.
Please use the Async API instead.

Tokens: 13815
Running open_url...
open_url returned 25 chars
Calling LLM with 10 messages, 32 tools
Tokens: 13901

In [27]

python

hooks = AgentHooks()
hooks.before_llm.append(lambda msgs, tools: print("calling LLM"))

In [28]

python

total_cost = {"tokens": 0}


@hooks.on_after_llm
def accumulate(response):
    total_cost["tokens"] += response.usage.get("total_tokens", 0)


agent.run("Do something complex")
print(f"Total tokens used: {total_cost['tokens']}")

Out

Calling LLM with 2 messages, 32 tools
Tokens: 12394
Running ask_user...
ask_user returned 427 chars
Calling LLM with 4 messages, 32 tools
Tokens: 12474
Total tokens used: 0

In [29]

python

import time

last_call = {"time": 0.0}


@hooks.on_before_llm
def rate_limit(messages, tools):
    elapsed = time.time() - last_call["time"]
    if elapsed < 1.0:
        time.sleep(1.0 - elapsed)
    last_call["time"] = time.time()

In [30]

python

@hooks.on_after_tool
def filter_pii(name, result):
    if "email" in result.output.lower():
        print(f"Warning: {name} output may contain PII")

In [31]

python

BLOCKED_TOOLS = {"code_execution", "workspace_files"}


@hooks.on_before_tool
def block_dangerous_tools(name, arguments):
    if name in BLOCKED_TOOLS:
        raise PermissionError(f"Tool {name} is blocked by policy")

In [3]

python

from shipit_agent import Agent

agent = Agent.with_builtins(
    llm=llm,
    parallel_tool_execution=True,  # run tools concurrently
)

# result = agent.run("Search for Python 3.13 release notes and today's weather in NYC")
for event in agent.stream(
    "Search for Python 3.13 release notes and today's weather in NYC"
):
    if event.type == "tool_called":
        print(f"Tool called: {event.payload.get('arguments', {})}")
    elif event.type == "tool_completed":
        print(f"Tool done: {event.payload.get('output', '')[:200]}")
    elif event.type == "reasoning_completed":
        print(f"Reasoning: {event.payload.get('content', '')[:200]}")
    elif event.type == "run_completed":
        print(f"Final: {event.payload.get('content', '')[:500]}")
    else:
        print(f"[{event.type}] {event.message}")

Out

[run_started] Agent run started
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: The user wants two pieces of info: Python 3.13 release notes and today's weather in NYC.

We need to fetch up-to-date info. Use web_search for Python 3.13 release notes (maybe "Python 3.13 release not
Tool called: {'query': 'Python 3.13 release notes', 'max_results': 5}
Tool done: [1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains the new features in Python 3.13 , compared to 3.12
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need Python 3.13 release notes and today's weather in NYC. We have search results for release notes. Need to fetch the release notes page. Also need weather. Likely use web search for weather NYC t
Tool called: {'max_results': 5, 'query': 'NYC weather today'}
Tool done: [1] Manhattan, New York City, New York - The Weather Channel
Today's and tonight's Manhattan, New York City, New York weather forecast, weather conditions and Doppler radar from The Weather Channel an
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need to retrieve Python 3.13 release notes and today's weather in NYC. We have search results. Need to open likely sources for both. For Python, the official docs "What's New In Python 3.13". Let's
Tool called: {'url': 'https://docs.python.org/3/whatsnew/3.13.html', 'max_chars': 2000}
Tool done: index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한국어
Polish | polski
Brazilian Portuguese | Portuguê
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We have Python 3.13 release notes summary. Need also today's weather in NYC. Let's open a weather site to get current data. Use open_url on a reliable source like weather.com but need specific info. U
Tool called: {'url': 'https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf', 'max_chars': 1500}

open_url: Playwright fetch failed for https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf: HTTP 404 from https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf
open_url: urllib fetch failed for https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf: HTTP Error 404: Not Found

Tool done: Error: could not fetch https://weather.com/weather/today/l/Manhattan+New+York+City+New+York?canonicalCityId=4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf — playwright failed: HTTP 4
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: We need today's weather in NYC. Could use another source like weather.gov or AccuWeather or simple API. Let's open a more reliable URL. Use National Weather Service.
Tool called: {'url': 'https://forecast.weather.gov/MapClick.php?textField1=40.71&textField2=-74.01', 'max_chars': 2000}
Tool done: HOME
FORECAST 
PAST WEATHER 
SAFETY 
INFORMATION 
EDUCATION 
NEWS 
SEARCH 
ABOUT 
 
View Location Examples

Your local forecast office is

New York, NY
 
News Headlines
Learn All About Being a COCORAH
[step_started] LLM completion started
[reasoning_started] Model reasoning started
Reasoning: User asks: "Search for Python 3.13 release notes and today's weather in NYC". Need to provide both pieces of info. We have fetched Python release notes page with summary. Could extract key highlights.
Final: **Python 3.13 – Release Notes (high‑level highlights)**  
Source: *What’s New in Python 3.13* (Python documentation)  

- **Release date:** 7 Oct 2024.  
- **Major language changes**
  * New interactive interpreter (a refreshed REPL) with improved UI.
  * Experimental **free‑threaded mode** (PEP 703) – allows running Python code without the GIL for certain workloads.
  * **Just‑In‑Time (JIT) compiler** (PEP 744) – optional performance boost for hot code paths.
  * `locals()` now has defined sema

In [13]

python

from shipit_agent import Agent, FileTraceStore

agent = Agent(
    llm=llm,
    max_iterations=10,
    trace_store=FileTraceStore(root_dir=".shipit_traces"),
    trace_id="run-2026-04-09-alice",
)

In [7]

python

from IPython.display import Markdown, display


def markdown(text):
    display(Markdown(text))

In [ ]

python

for event in agent.stream(
    "Search for Python 3.13 release notes and today's weather in NYC"
):
    if event.type == "tool_called":
        print("Tool called:")
        markdown(f"**Tool Called:** `{event.payload.get('arguments', {})}`")
    elif event.type == "tool_completed":
        print("Tool done:")
        markdown(
            f"**Tool Output:**\n```\n{event.payload.get('output', '')[:1000]}\n```"
        )
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
        markdown(event.payload.get("content", ""))
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to do web searches. Let's search for "Python 3.13 release notes".

Tool called:

Tool Called: {'query': 'Python 3.13 release notes', 'max_results': 5}

Tool done:

Tool Output:

[1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains the new features in Python 3.13 , compared to 3.12. Python 3.13 was released on October 7, 2024. For full details, see the changelog.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fdocs.python.org%2F3%2Fwhatsnew%2F3.13.html&rut=1efa42f765ec1a7db1a6eeec0a7b9eab118df2821166b88c237d1441e91fed72

[2] Python 3.13 - Release History, LTS & EOL - VersionLog
Python 3.13 Release Notes Python 3.13.0, released on October 7, 2024, introduces major advancements in performance, concurrency, and developer experience.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fversionlog.com%2Fpython%2F3.13%2F&rut=a344c1429deb5f549336315491ff0bac94071ac066ce29d28af00b6eae1eb4b2

[3] cpython/Doc/whatsnew/3.13.rst at main · python/cpython
Summary -- Release Highlights Python 3.13 is a stable release of the Python programming language, with a mix of changes to the language, th

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need both Python 3.13 release notes and today's weather in NYC. Need to fetch weather info. We'll search for NYC weather today.

Tool called:

Tool Called: {'max_results': 5, 'query': 'New York City weather today'}

Tool done:

Tool Output:

[1] New York, NY Weather Forecast | AccuWeather
New York , NY Weather Forecast, with current conditions, wind, air quality, and what to expect for the next 3 days.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.accuweather.com%2Fen%2Fus%2Fnew%2Dyork%2F10021%2Fweather%2Dforecast%2F349727&rut=d12472aa45963856a03b496d2712338ee0d0e68923cc9e83f8d1cf2e302140e1

[2] 10-day weather forecast for Manhattan, New York City, New York, United ...
Be prepared with the most accurate 10-day forecast for Manhattan, New York City , New York , United States with highs, lows, chance of precipitation from The Weather Channel and Weather .com
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fweather.com%2Fen%2DTO%2Fweather%2Ftenday%2Fl%2FManhattan%2BNew%2BYork%2BCity%2BNew%2BYork%2BUnited%2BStates%3FplaceId%3D4020d3357937ae0b52673ba25c8906efd1d8011774f10ff04551f34eadf31fdf&rut=e9b5a9661ca09a5e9522694395ef367875be8734ffaaa5b4b52a6a0abf39e7d2

[3] 7-Day Forecast 40.71N 73.98W - National Weather Service
Hazardous We

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We need to provide Python 3.13 release notes and today's weather in NYC.

We have search results. Need to open relevant pages to extract info.

Open Python 3.13 what's new page.

Tool called:

Tool Called: {'url': 'https://docs.python.org/3/whatsnew/3.13.html'}

Tool done:

Tool Output:

index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한국어
Polish | polski
Brazilian Portuguese | Português brasileiro
Romanian | Românește
Turkish | Türkçe
Simplified Chinese | 简体中文
Traditional Chinese | 繁體中文
dev (3.15)
3.14.4
3.13
3.12
3.11
3.10
3.9
3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
3.0
2.7
2.6
 3.14.4 Documentation » What’s New in Python » What’s New In Python 3.13
  |
… 22 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

Tool called:

Tool Called: {'url': 'https://wttr.in/New+York?format=3'}

open_url: Playwright fetch failed for https://wttr.in/New+York?format=3: Page.goto: net::ERR_ABORTED at https://wttr.in/New+York?format=3
Call log:
  - navigating to "https://wttr.in/New+York?format=3", waiting until "domcontentloaded"

Tool done:

Tool Output:

new york: ☀️ +12°C

[step_started] LLM completion started

Python 3.13 – Release notes (summary)

Release date: 7 Oct 2024 (official Python documentation)
Key highlights:
- New interactive interpreter with colored tracebacks.
- PEP 703: experimental free‑threaded mode (optional GIL‑disable).
- PEP 744: basic JIT compiler (disabled by default).
- locals() now has defined semantics when the returned mapping is mutated.
- Type parameters can now have default values.
- Standard‑library changes: new PythonFinalizationError, deprecations in argparse, base64.z85encode/decode, copy.replace, default dbm.sqlite3 backend, new os timer‑fd helpers, CLI for random, etc.
- Security: ssl.create_default_context() now enforces VERIFY_X509_PARTIAL_CHAIN.

(Full details are in the official “What’s New in Python 3.13” page: https://docs.python.org/3/whatsnew/3.13.html)

Today's weather in New York City (as of the time of the query)

Condition: Sunny ☀️
Temperature: +12 °C

Source: https://wttr.in/New+York?format=3

Let me know if you need more detail on any of these topics!

In [10]

python

for event in agent.stream(
    "Search for Python 3.14.4 release notes and today's weather in Warsaw"
):
    if event.type == "tool_called":
        print("Tool called:")
        markdown(f"**Tool Called:** `{event.payload.get('arguments', {})}`")
    elif event.type == "tool_completed":
        print("Tool done:")
        markdown(
            f"**Tool Output:**\n```\n{event.payload.get('output', '')[:1000]}\n```"
        )
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
        markdown(event.payload.get("content", ""))
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

Tool called:

Tool Called: {'query': 'Python 3.14.4 release notes', 'max_results': 5}

Tool done:

Tool Output:

[1] Python Release Python 3.14.4 | Python.org
Python 3.14.4 is the fourth maintenance release of 3.14, containing around 337 bugfixes, build improvements and documentation changes since 3.14.3. Major new features of the 3.14 series, compared to 3.13
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.python.org%2Fdownloads%2Frelease%2Fpython%2D3144%2F&rut=4b57433242b8a3a3eaf4cad8dec0ed94bd8ef4d04cf800e1b21d98a53c14c9bb

[2] Changelog — Python 3.14.3 documentation
This avoids issues with upgrading Python while it is running. ( Note that such 'in-place' upgrades are not tested.) The tracker remains compatible with subprocesses that use new protocol (that is, subprocesses using Python 3.13.10, 3.14.1 and 3.15). gh-142214: Fix two regressions in dataclasses in Python 3.14.1 related to annotations.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fdocs.python.org%2F3%2Fwhatsnew%2Fchangelog.html&rut=5053dd15419ac522349e2e7b5efefbd7c2a7ce91a4c44a9a6458aa9a91282d64

[3] Python 3.15.0a8, 3.14.4 and 3.13.

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

The user asks: "Search for Python 3.14.4 release notes and today's weather in Warsaw". Need to provide both pieces of information.

We have search results for Python 3.14.4 release notes. Let's open the first result.

Tool called:

Tool Called: {'url': 'https://www.python.org/downloads/release/python-3144/'}

Tool done:

Tool Output:

Skip to content
Python
PSF
Docs
PyPI
Jobs
Community
Donate
 
Search This Site
 GO
Socialize
About
 
Downloads
 
Documentation
 
Community
 
Success Stories
 
News
 
Events
Python 3.14.4

Release date: April 7, 2026

This is fourth maintenance release of Python 3.14

Python 3.14.4 is the fourth maintenance release of 3.14, containing around 337 bugfixes, build improvements and documentation changes since 3.14.3.

Major new features of the 3.14 series, compared to 3.13

Some of the major new features and changes in Python 3.14 are:

New features
… 7 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

Tool called:

Tool Called: {'max_results': 5, 'query': 'Warsaw weather today'}

Tool done:

Tool Output:

[1] Today's Weather in Warszawa - Hourly Forecast and Conditions
Get the latest hourly weather updates for Warszawa today . Detailed forecast including temperature, wind, rain, snow, and UV index. Stay informed about today's weather conditions in Warszawa.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.easeweather.com%2Feurope%2Fpoland%2Fmazovia%2Fwarszawa%2Ftoday&rut=3dedf7325907f92328e940c240a1553e2d7dd1d46f2891d49f75ef4cf9f46ae9

[2] Warsaw Weather Forecast
12 day Warsaw Weather Forecast. Live Weather Warnings, hourly weather updates. Accurate Warsaw weather today , forecast for sun, rain, wind and temperature.
URL: //duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.weather%2Dforecast.com%2Flocations%2FWarsaw%2Fforecasts%2Flatest&rut=f512cd7bae35b6d3a813f459fa24d6236565d45c9fdf79fb87b30da892ae2d30

[3] Warsaw, Masovia, Poland Hourly Weather | AccuWeather
Hourly weather forecast in Warsaw , Masovia, Poland. Check current conditions in Warsaw , Masovia, Poland with radar, hourly, and more.
U

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

Tool called:

Tool Called: {'url': 'https://weather.com/weather/today/l/Warsaw+Poland?canonicalCityId=7ed0ae4e25e1adb77367c6f70331d838'}

Tool done:

Tool Output:

Home
Today
Hourly
10 Day
Monthly
Radar
Video
Explore
Severe Weather
Hurricane Central
Tornado Central
Safety & Prep
Health & Wellness
Allergy Tracker
Air Quality Index
Cold & Flu Tracker
Skin
Lifestyle
Home & Garden
Pets & Animals
Recipes
Space & Skywatching
Travel & Outdoors
Product Reviews & Deals
Shop
Home
Today
Hourly
10 Day
Monthly
Radar
Video
Explore
Severe Weather
Hurricane Central
Tornado Central
Safety & Prep
Health & Wellness
… 18 more lines

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

We should ensure today's date (April 12, 2026). Provide weather details: temperature, conditions, wind, humidity, forecast.

Let's extract relevant info from weather page: In the snippet we see "As of 9:00 pm CEST 48° Fair". Also hourly shows now 46°, clear, etc. Provide summary.

Now craft answer.

Python 3.14.4 Release notes (released April 7 2026)

Overview
Python 3.14.4 is the fourth maintenance release of the 3.14 series. It bundles ≈ 337 bug‑fixes, assorted build improvements and documentation updates over 3.14.3.

Key highlights

Area	What’s new / changed
Free‑threaded interpreter	PEP 779 – free‑threaded Python is now officially supported.
Deferred annotation evaluation	PEP 649 – annotations are evaluated lazily, fixing long‑standing semantics issues.
Template string literals	PEP 750 – “t‑strings” give you a custom‑string‑processing syntax that mirrors f‑strings.
Multiple interpreters	PEP 734 – the stdlib now ships a multiple‑interpreter module.
Zstandard compression	PEP 784 – new `compression.zstd` module for Zstandard algorithm support.
Exception syntax	PEP 758 – `except`/`except*` can omit the surrounding brackets.
Debugger interface	PEP 768 – a zero‑overhead external debugger interface for CPython.
UUID extensions	UUID v6‑v8 added; generation of v3‑v5 now up to 40 % faster.
Return/break/continue in finally	PEP 765 disallows these statements from exiting a `finally` block.
C‑API improvements	PEP 741 – revamped C‑API for configuring Python.
New interpreter type	An optional, high‑performance interpreter for newer compilers (opt‑in).
Error‑message upgrades	More informative tracebacks and diagnostics.
Built‑in HMAC	Formal‑verification via the HACL* project.
Async process inspector	New CLI to inspect running Python processes using asynchronous tasks.
Remote pdb	`pdb` can now attach remotely to a live process.
Build changes	• No longer provides PGP signatures – Sigstore is recommended (PEP 761). • macOS/Windows binaries now include an experimental JIT. • Official Android binary releases are now available.
Installer shift	The traditional Windows installer is being replaced by a new Python Install Manager (available via Windows Store or direct download).

For a full, line‑by‑line changelog see the “Full Changelog” section on the release page or the “What’s new in Python 3.14” docs.

Warsaw (Poland) weather – 12 April 2026

Parameter	Value (as of the latest observation ≈ 21:00 CEST)
Temperature	48 °F (≈ 9 °C) – Fair conditions
Feels like	45 °F (≈ 7 °C)
Current condition	Clear skies
Wind	7 mph (≈ 11 km/h) from ESE
Humidity	45 %
… 16 more lines

In [17]

python

from shipit_agent.tools import WebSearchTool, OpenURLTool

agent = Agent(
    llm=llm,
    tools=[WebSearchTool(), OpenURLTool()],
    max_iterations=10,
    trace_store=FileTraceStore(root_dir=".shipit_traces"),
    trace_id="run-2026-04-09-alice",
)

In [16]

python

for event in agent.stream("Search for new about shipit agent and what is shipiit ai"):
    if event.type == "tool_called":
        print("Tool called:")
    elif event.type == "tool_completed":
        print("Tool done:")
    elif event.type == "reasoning_completed":
        print("Reasoning: ")
    elif event.type == "run_completed":
        markdown(event.payload.get("content", ""))
    else:
        markdown(f"**[{event.type}]** {event.message}")

Out

[run_started] Agent run started

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning: 
Tool called:
Tool done:

[step_started] LLM completion started

[reasoning_started] Model reasoning started

Reasoning:

ShipIt AI – Quick‑look (what it is)

Feature	Details
Core idea	An “AI‑developer‑workspace” that turns a plain‑English request into a full software‑delivery pipeline: parse → implement → review → fix → test → PR → ship.
AI engines	Supports up to four different generators (Claude, OpenAI Codex, Google Gemini, and a custom “ShipIt CLI” that can route to Bedrock, Ollama, Groq, Together, OpenRouter, etc.) and > 100 models.
Repo‑aware editing	The agent reads the entire repository, validates paths, and prevents out‑of‑scope edits.
Visible delivery loop	Every step (code generation, auto‑review, test run, commit) is shown in a UI/graph so the developer can see what the agent is doing, not just the final output.
Multi‑messenger integration	The UI can spin up Slack, Discord, and Telegram bots that listen for commands (e.g., “@shipit run the test suite”) and reply with live status updates.
Built‑in security / QA	Runs static analysis (Bandit, Semgrep), secret detection, license compliance, container scans, and an AI‑driven penetration‑testing suite (OWASP 2025 coverage).
Extensible workflow	30+ step‑type templates, drag‑and‑drop visual pipeline builder, 17+ MCP integrations (GitHub, GitLab, Jira, DBs, Playwright, etc.).
Public‑facing pages	• Product landing: https://shipiit.com/ • LinkedIn announcement (4 days ago): https://www.linkedin.com/posts/shipiit_shipit-a-powerful-ai-coding-agent-with-activity-7447720216366682112-1Z6I • GitHub “ShipIt‑AI” graph project: https://github.com/ship-it-ops/ShipIt-AI • “ShipIt Clean” code‑analysis spinoff: https://shipitclean.com/ • Devpost hackathon entry (2018): https://devpost.com/software/shipit

Recent news & activity (the “new about ShipIt agent”)

Date (≈)	Source	Highlight
4 days ago	LinkedIn post by ShipIt (35 followers)	Announces a “powerful AI coding agent with a beautiful modern UI” that works across Slack, Discord and Telegram. Shows a concrete workflow where a user asks the bot to run a test suite and receives a live‑streamed result. Includes a wait‑list link to early‑access the product.
2024‑03 (approx.)	ShipIt product site (shipiit.com)	Updated landing page now shows four AI engines, 100+ models, 30+ visual workflow steps and 17+ integrations. Emphasises “repo‑aware before editing”, “visible delivery graph”, and “operator‑first product”.
2024‑02	ShipIt‑Clean website launch	Introduces a separate service that runs multiple independent AI agents to audit a codebase (feature completeness, architecture, performance, developer experience, scalability) and output a prioritized Opportunity Report.
2024‑02‑20 (GitHub commit)	GitHub – ship‑it‑ops/ShipIt‑AI	Open‑source project that maps a software ecosystem into a Neo4j knowledge graph. Connectors pull data from GitHub, Kubernetes, etc., exposing a queryable graph for downstream AI agents (Claude, Gemini, etc.).
2023‑12 (Devpost)	Devpost “ShipIt” hackathon project	Earlier incarnation of “ShipIt” that combined a smart contract on Ethereum with a ML model predicting project completion time. Demonstrates the team’s long‑standing interest in AI‑assisted project delivery.

TL;DR Summary

ShipIt AI is an end‑to‑end AI‑powered development platform that turns natural‑language prompts into a full CI/CD‑style workflow, complete with code generation, automated review, testing, security scanning, and PR creation.
It runs multiple large‑model back‑ends (Claude, Codex, Gemini, custom CLI) and is tightly integrated with your code repository, so … output trimmed

Load Trace

Here's how the tracing works and how to view traces nicely like LangSmith/LangGraph

In [18]

python

store = FileTraceStore(root_dir=".shipit_traces")
trace = store.load("run-2026-04-09-alice")

# Pretty print like LangGraph/LangSmith
for i, event in enumerate(trace.events):
    indent = (
        "  "
        if event.type in ("tool_called", "tool_completed", "reasoning_completed")
        else ""
    )

    if event.type == "run_started":
        print(f"🟢 Run Started: {event.payload.get('prompt', '')}")
    elif event.type == "step_started":
        print(f"\n📍 Step {event.payload.get('iteration', '?')}")
    elif event.type == "reasoning_completed":
        print(f"  💭 Reasoning: {event.payload.get('content', '')[:150]}...")
    elif event.type == "tool_called":
        print(f"  🔧 → {event.payload.get('arguments', {})}")
    elif event.type == "tool_completed":
        print(f"  ✅ ← {event.payload.get('output', '')[:150]}")
    elif event.type == "tool_failed":
        print(f"  ❌ Failed: {event.payload.get('error', '')}")
    elif event.type == "run_completed":
        usage = event.payload.get("usage", {})
        print(f"\n🏁 Done — tokens: {usage.get('total_tokens', 0)}")
        print(f"   Output: {event.payload.get('content', '')[:300]}")
    else:
        print(f"  [{event.type}] {event.message}")

Out

🟢 Run Started: Search for Python 3.13 release notes and today's weather in NYC

📍 Step 1
  [reasoning_started] Model reasoning started
  💭 Reasoning: The user asks: "Search for Python 3.13 release notes and today's weather in NYC". So need to provide two pieces: Python 3.13 release notes, and today'...
  🔧 → {'query': 'Python 3.13 release notes', 'max_results': 5}
  ✅ ← [1] What's New In Python 3.13 — Python 3.14.3 documentation
What's New In Python 3.13 ¶ Editors: Adam Turner and Thomas Wouters This article explains 

📍 Step 2
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need both Python 3.13 release notes and today's weather in NYC. Need to fetch weather info. We'll search for NYC weather today....
  🔧 → {'max_results': 5, 'query': 'New York City weather today'}
  ✅ ← [1] New York, NY Weather Forecast | AccuWeather
New York , NY Weather Forecast, with current conditions, wind, air quality, and what to expect for the

📍 Step 3
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need to provide Python 3.13 release notes and today's weather in NYC.

We have search results. Need to open relevant pages to extract info.

Open P...
  🔧 → {'url': 'https://docs.python.org/3/whatsnew/3.13.html'}
  ✅ ← index
modules |
next |
previous |
 Python » 
Greek | Ελληνικά
English
Spanish | español
French | français
Italian | italiano
Japanese | 日本語
Korean | 한

📍 Step 4
  [reasoning_started] Model reasoning started
  💭 Reasoning: We need to provide both Python 3.13 release notes summary and today's weather in NYC. We have release notes page fetched. Need to extract key points (...
… 420 more lines

In [19]

python

from shipit_agent import Agent

# ToolSearchTool is included in Agent.with_builtins automatically
agent = Agent.with_builtins(llm=llm)

result = agent.run(
    "Use tool_search to find the right tool for fetching a specific URL, "
    "then call that tool on https://example.com"
)

In [21]

python

markdown(f"**Final Result:**\n```\n{result.output}\n```")

Out

Final Result:

Here’s the content fetched from **https://example.com** (truncated to the first 1 000 characters):

Example Domain

This domain is for use in documentation examples without needing permission. Avoid use in operations.

Learn more

In [4]

python

# ToolSearchTool is included in Agent.with_builtins automatically
from shipit_agent import Agent

agent = Agent.with_builtins(llm=llm)

In [5]

python

from pydantic import BaseModel
from shipit_agent import Agent


class MovieReview(BaseModel):
    title: str
    rating: float
    summary: str
    earning: int
    pros: list[str]
    cons: list[str]
    casts: list[str]


result = agent.run("Review The Dhurandhar: The Revenge", output_schema=MovieReview)

review = result.parsed  # MovieReview instance
review.title  # "The Dhurandhar: The Revenge"
review.rating  # 9.5
review.earning  # 1000000
review.pros  # ["Groundbreaking visuals", ...]
review.cons  # ["Weak storyline", ...]
review.casts  # ["Actor 1", "Actor 2", ...]
type(review)  # <class 'MovieReview'>

Out

__main__.MovieReview

In [9]

python

print(f"Title: {review.title}")
print(f"Rating: {review.rating}")
print(f"Earning: ${review.earning}")
print("Pros:")
for p in review.pros:
    print(f"- {p}")
print("Cons:")
for c in review.cons:
    print(f"- {c}")
print("Casts:")
for cast in review.casts:
    print(f"- {cast}")

Out

Title: Dhurandhar: The Revenge
Rating: 3.5
Earning: $16822200000
Pros:
- Strong lead performance by Ranveer Singh
- Impressive supporting cast (Arjun Rampal, Sanjay Dutt, R. Madhavan, Sara Arjun)
- High‑production values and expansive action set‑pieces
- Well‑received soundtrack and technical craftsmanship
- Record‑breaking box‑office performance
Cons:
- Overly long (nearly 4 hours) leading to pacing issues
- Graphic violence and gore that alienated many viewers
- Criticized for overt nationalist messaging and propaganda
- Mixed critical scores (35 % Tomatometer on Rotten Tomatoes)
- Limited appeal outside the core Indian market due to political tone
Casts:
- Ranveer Singh
- Arjun Rampal
- Sanjay Dutt
- R. Madhavan
- Sara Arjun
- Rakesh Bedi
- Gaurav Gera
- Danish Pandor
- Manav Gohil

In [28]

python

resultreview = result.parsed

if review is None:
    print("⚠️ No structured output, generating fallback...")

    review = MovieReview(
        title="The Dhurandhar: The Revenge",
        rating=6.5,
        summary="A fictional action drama centered on revenge with intense sequences.",
        earning=50000000,
        pros=["Action scenes", "Lead performance"],
        cons=["Predictable story", "Weak pacing"],
        casts=["Actor A", "Actor B"],
    )

Out

⚠️ No structured output, generating fallback...

In [24]

python

review = result.parsed

In [25]

python

review

In [26]

python

from pydantic import BaseModel
from shipit_agent import Agent


class MovieReview(BaseModel):
    title: str
    rating: float
    summary: str
    pros: list[str]
    cons: list[str]


agent = Agent.with_builtins(llm=llm)
result = agent.run("Review The Matrix", output_schema=MovieReview)

review = result.parsed  # MovieReview instance
review.title  # "The Matrix"
review.rating  # 9.5
review.pros  # ["Groundbreaking visuals", ...]
type(review)  # <class 'MovieReview'>

Out

__main__.MovieReview