Notebooks · First agents · 02

Multi-Tool Agent

Name: SHIPIT Agent
Author: SHIPIT

This notebook wires multiple tools into one agent so you can test tool selection and tool results.

Open in Colab View on GitHub 3 code cells · 2 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 [1]

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_demo_agent, build_llm_from_env

In [2]

python

llm = build_llm_from_env("bedrock")

agent = build_demo_agent(
    llm=llm,
    workspace_root=str(ROOT / ".shipit_notebooks" / "multi_tools"),
)

sorted(tool.name for tool in agent.tools)[:20]

Out

['add_numbers',
 'ask_user',
 'build_artifact',
 'build_prompt',
 'confluence',
 'custom_api',
 'decision_matrix',
 'decompose_problem',
 'gmail_search',
 'google_calendar',
 'google_drive',
 'human_review',
 'jira',
 'linear',
 'memory',
 'notion',
 'open_url',
 'plan_task',
 'playwright_browse',
 'project_context']

In [3]

python

from IPython.display import Markdown, display

result = agent.run(
    "Use project_context and workspace_files if useful, then summarize what this agent workspace is set up to do."
)
display(Markdown(result.output))
[(item.name, item.output[:120]) for item in result.tool_results]

Out

Summary of this agent’s workspace configuration

Aspect	Details
Core runtime	Shipit – a general‑purpose, tool‑driven execution environment that emphasizes accurate, execution‑oriented responses.
Primary design goals	• Provide factual, up‑to‑date answers. • Use external tools (web search, browsing, code execution, APIs, etc.) whenever they improve correctness or freshness. • Keep reasoning transparent and structured (plans, decompositions, evidence synthesis).
Toolset available	• Web: `web_search`, `open_url`, `playwright_browse` for fresh public data. • Communication: `gmail_search`, `google_calendar`, `google_drive`, `slack`, `linear`, `jira`, `notion`, `confluence` for internal org data. • Data handling: `workspace_files` (read/write), `run_code` (Python/Bash), `build_artifact` (rich visual output). • Decision & Planning: `plan_task`, `decompose_problem`, `synthesize_evidence`, `decision_matrix`, `verify_output`. • Memory: `memory` for persisting facts across a session. • Utility: `add_numbers`, `project_context`, `custom_api`, `sub_agent`.
Preferred workflow	1. Clarify user intent (ask only when needed). 2. Plan complex tasks (`plan_task`/`decompose_problem`). 3. Gather evidence using the most appropriate tool (prefer a single, focused tool). 4. Process data with `run_code` or built‑in utilities. 5. Validate results (`verify_output`). 6. Present output directly, or as a rich artifact (`build_artifact`).
Human‑in‑the‑loop	Uses `ask_user` for clarification and `human_review` before any irreversible action (e.g., sending email, creating tickets, publishing docs).
Memory handling	Facts are stored compactly via `memory` and retrieved before re‑asking the user.
Current file system state	• `memory.json` – empty list (no stored facts yet). • `sessions` – directory for session‑specific data. • `traces` – directory for execution traces.
Constraints	• No fabricated data – always ground answers in evidence or clearly label assumptions. • Keep responses concise, structured, and easy to verify. • Use the smallest, most specific tool for each sub‑task.

What this workspace can do for you

Retrieve up‑to‑date web information or internal corporate data via the appropriate connectors.
Run code, manipulate files, and generate downloadable artifacts.
Plan and execute multi‑step processes with clear checkpoints and risk awareness.
Store reusable knowledge across the session and recall it on demand.

Feel free to ask for any specific operation—e.g., a web lookup, a file creation, a calendar query, or a full end‑to‑end workflow—and the agent will orchestrate the necessary tools to deliver a reliable result.

[('project_context',
  'This workspace uses SHIPIT Agent for tool-using workflows. Prefer grounded answers, explain tradeoffs clearly, and use t'),
 ('workspace_files', 'memory.json\nsessions\ntraces'),
 ('workspace_files', '[]')]

Multi-Tool Agent

This notebook wires multiple tools into one agent so you can test tool selection and tool results.

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

In [1]

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_demo_agent, build_llm_from_env

In [2]

python

llm = build_llm_from_env("bedrock")

agent = build_demo_agent(
    llm=llm,
    workspace_root=str(ROOT / ".shipit_notebooks" / "multi_tools"),
)

sorted(tool.name for tool in agent.tools)[:20]

Out

['add_numbers',
 'ask_user',
 'build_artifact',
 'build_prompt',
 'confluence',
 'custom_api',
 'decision_matrix',
 'decompose_problem',
 'gmail_search',
 'google_calendar',
 'google_drive',
 'human_review',
 'jira',
 'linear',
 'memory',
 'notion',
 'open_url',
 'plan_task',
 'playwright_browse',
 'project_context']

In [3]

python

from IPython.display import Markdown, display

result = agent.run(
    "Use project_context and workspace_files if useful, then summarize what this agent workspace is set up to do."
)
display(Markdown(result.output))
[(item.name, item.output[:120]) for item in result.tool_results]

Out

Summary of this agent’s workspace configuration

Aspect	Details
Core runtime	Shipit – a general‑purpose, tool‑driven execution environment that emphasizes accurate, execution‑oriented responses.
Primary design goals	• Provide factual, up‑to‑date answers. • Use external tools (web search, browsing, code execution, APIs, etc.) whenever they improve correctness or freshness. • Keep reasoning transparent and structured (plans, decompositions, evidence synthesis).
Toolset available	• Web: `web_search`, `open_url`, `playwright_browse` for fresh public data. • Communication: `gmail_search`, `google_calendar`, `google_drive`, `slack`, `linear`, `jira`, `notion`, `confluence` for internal org data. • Data handling: `workspace_files` (read/write), `run_code` (Python/Bash), `build_artifact` (rich visual output). • Decision & Planning: `plan_task`, `decompose_problem`, `synthesize_evidence`, `decision_matrix`, `verify_output`. • Memory: `memory` for persisting facts across a session. • Utility: `add_numbers`, `project_context`, `custom_api`, `sub_agent`.
Preferred workflow	1. Clarify user intent (ask only when needed). 2. Plan complex tasks (`plan_task`/`decompose_problem`). 3. Gather evidence using the most appropriate tool (prefer a single, focused tool). 4. Process data with `run_code` or built‑in utilities. 5. Validate results (`verify_output`). 6. Present output directly, or as a rich artifact (`build_artifact`).
Human‑in‑the‑loop	Uses `ask_user` for clarification and `human_review` before any irreversible action (e.g., sending email, creating tickets, publishing docs).
Memory handling	Facts are stored compactly via `memory` and retrieved before re‑asking the user.
Current file system state	• `memory.json` – empty list (no stored facts yet). • `sessions` – directory for session‑specific data. • `traces` – directory for execution traces.
Constraints	• No fabricated data – always ground answers in evidence or clearly label assumptions. • Keep responses concise, structured, and easy to verify. • Use the smallest, most specific tool for each sub‑task.

What this workspace can do for you

Retrieve up‑to‑date web information or internal corporate data via the appropriate connectors.
Run code, manipulate files, and generate downloadable artifacts.
Plan and execute multi‑step processes with clear checkpoints and risk awareness.
Store reusable knowledge across the session and recall it on demand.

[('project_context',
  'This workspace uses SHIPIT Agent for tool-using workflows. Prefer grounded answers, explain tradeoffs clearly, and use t'),
 ('workspace_files', 'memory.json\nsessions\ntraces'),
 ('workspace_files', '[]')]