Notebooks
Learn it by running it.
Jupyter notebooks, cell by cell, with the outputs they saved. Follow the guides in order, or jump to a single feature. Read them here, open them in Colab, or clone the repo and run them.
- notebooks
- 92
- in-depth guides
- 13
- code cells
- 965
- match 3.0's API
- 92
guides/06_rag.ipynbPython 3 · running
In [*]
rag = RAG.default(embedder=CallableEmbedder(titan, dimension=256))for path in sorted(Path("kb").glob("*.md")): rag.index_file(str(path))print(rag.list_sources())
In [ ]
agent = Agent(llm=gemma, rag=rag)result = agent.run( "If I cancel a monthly Team plan after 45 days, what do I get back?")
In [ ]
for s in result.rag_sources: print(f"[{s.index}] {s.source}")
Learning path
13 guides, in order.
13 of 13 written · each run end to end on shipit-agent 3.0, on two models
- Guide 01ReadySetup, .env and every model provider17 cells · 16 with output
- Guide 02ReadyThe Agent: run(), stream(), results and events21 cells · 21 with output
- Guide 03ReadyTools: your functions, classes and the built-ins17 cells · 15 with output
- Guide 04ReadyMCP servers15 cells · 13 with output
- Guide 05ReadyConnections and OAuth17 cells · 15 with output
- Guide 06ReadyRAG: search your documents, cite the passages22 cells · 20 with output
- Guide 07ReadySessions and memory15 cells · 13 with output
- Guide 08ReadyStructured output12 cells · 11 with output
- Guide 09ReadyPermissions, hooks and the sandbox15 cells · 14 with output
- Guide 10ReadySub-agents and the verifier10 cells · 8 with output
- Guide 11ReadyDeep agents and autopilot10 cells · 8 with output
- Guide 12ReadyCost, tracing, serving and output13 cells · 11 with output
- Guide 13ReadyChat history14 cells · 12 with output
92 notebooks
In-depth guides
Long walkthroughs for 3.0, run end to end on Haiku 5.5 (Anthropic) and Gemma 4 (Bedrock).
- Guide 01In-depth guidesSetup, .env and every model providershipit-agent runs the same agent loop on any model. This guide sets up a project, shows how the environment picks a provider, and runs the same question on Haiku 5.5 on the Anthropic API and on Gemma 4 31B on AWS…17 code cells16 with outputAPI checked
- Guide 02In-depth guidesThe Agent: run(), stream(), results and eventsEvery output below is real: the notebook was run end to end on Haiku 5.5 (Anthropic) and Gemma 4 31B.21 code cells21 with outputAPI checked
- Guide 03In-depth guidesTools: your functions, classes and the built-insA tool is a name, a description, a JSON Schema for its arguments, and code that runs. The model decides when to call one; shipit runs it, handles its errors, and feeds the result back.17 code cells15 with outputAPI checked
- Guide 04In-depth guidesMCP serversThe Model Context Protocol (MCP) is a standard way to give an agent tools that live in another process: a local program that talks over stdin and stdout, or a hosted server over HTTP. shipit connects to either…15 code cells13 with outputAPI checked
- Guide 05In-depth guidesConnections and OAuthA connection is a service your agent acts in on someone's behalf: Linear, GitHub, Slack, Notion. In shipit every connector is an MCP server from a catalog. Some read a key from the environment; the hosted ones use…17 code cells15 with outputAPI checked
- Guide 06In-depth guidesRAG: search your documents, cite the passagesRetrieval-augmented generation answers from your documents. shipit's RAG indexes text and files into chunks, searches them with vectors and keywords together, and gives an agent search tools plus the instructions to…22 code cells20 with outputAPI checked
- Guide 07In-depth guidesSessions and memoryA session is one conversation that continues: each run sees what came before, even from a new process. Memory is what lasts beyond a conversation: facts about a user or a project. shipit has both, and a way for the…15 code cells13 with outputAPI checked
- Guide 08In-depth guidesStructured outputAsk for a shape, get it back validated. Pass a Pydantic model (or a JSON Schema) and the answer arrives as a typed object. An answer that doesn't fit goes back to the model with the validation error, in the same…12 code cells11 with outputAPI checked
- Guide 09In-depth guidesPermissions, hooks and the sandboxAn agent that can edit files and run commands needs limits. shipit has three layers:15 code cells14 with outputAPI checked
- Guide 10In-depth guidesSub-agents and the verifierOne agent can hand work to others: a specialist with its own prompt, tools and model, or several helpers in parallel. And a second, cheap model can watch the first: veto a tool call that's off-task, and refuse to let a…10 code cells8 with outputAPI checked
- Guide 11In-depth guidesDeep agents and autopilotFor work that takes many steps, shipit has agents with structure built in:10 code cells8 with outputAPI checked
- Guide 12In-depth guidesCost, tracing, serving and outputRunning agents for real means knowing what each run cost, keeping a record of what happened, showing it to people, and sending it over the web. This guide covers all four.13 code cells11 with outputAPI checked
- Guide 13In-depth guidesChat historyA chat app keeps the conversation somewhere (your database, a session store) and hands it to the agent on each request. This guide covers every way history gets into and out of a shipit agent.14 code cells12 with outputAPI checked
Earlier notebooks
Shorter notebooks from earlier releases, one feature each. Their code is checked against 3.0; their saved outputs are from the release they were written for.
- 01First agentsBasic Agent Without ToolsUse this notebook to test the smallest possible Agent setup.13 code cells7 with outputAPI checked
- 02First agentsMulti-Tool AgentThis notebook wires multiple tools into one agent so you can test tool selection and tool results.3 code cells2 with outputAPI checked
- 03First agentsSession History And Seeded Chat HistoryThis notebook shows both ways to preserve context:4 code cells3 with outputAPI checked
- 04First agentsStreaming Events And PacketsUse this notebook to inspect the real-time runtime stream, collected event packets, and frontend-ready packet transports.9 code cells4 with outputAPI checked
- 05First agentsAgent With MCPThis notebook shows how to attach remote MCP servers and combine them with local tools.2 code cellsAPI checked
- 06First agentsGmail And Connector ToolsThis notebook shows how to attach Gmail and other connector tools through the credential store.2 code cellsAPI checked
- 07First agentsHuman-In-The-Loop ReviewUse this notebook to test HumanReviewTool and the runtime's interactive review event flow.2 code cellsAPI checked
- 08First agentsAsk User FlowUse this notebook to test AskUserTool and the runtime's clarification request pattern.2 code cellsAPI checked
- 09First agentsVisible Reasoning And Thinking ToolsThis notebook demonstrates the safe, inspectable reasoning layer in SHIPIT Agent.5 code cellsAPI checked
- 10Output, pipelines and memoryParsers & Structured OutputGet typed, validated responses from LLMs. SHIPIT supports Pydantic models, JSON schemas, regex patterns, and markdown extraction — all as standalone parsers or as a one-parameter addition to agent.run().10 code cells8 with outputAPI checked
- 11Output, pipelines and memoryPipeline CompositionChain agents and functions together into deterministic workflows. Cleaner than LangChain's LCEL.6 code cells5 with outputAPI checked
- 12Output, pipelines and memoryAgent TeamsDynamic, LLM-routed multi-agent collaboration. Define agents with roles, and a coordinator decides who works, in what order, when to loop back. No manual graph wiring needed.4 code cells3 with outputAPI checked
- 13Output, pipelines and memoryAdvanced Memory SystemThree types of memory working together: conversation, semantic, and entity memory.5 code cells4 with outputAPI checked
- 14Deep agentsDeep AgentsThe features that put SHIPIT beyond LangChain. Autonomous, self-directing, self-correcting agents.8 code cells7 with outputAPI checked
- 15Deep agentsAgent BenchmarkingSystematically test your agents with expected outputs, tool usage, and content checks. Built-in evaluation framework — nothing like this in LangChain.4 code cells3 with outputAPI checked
- 16Deep agentsDeep Agents: Real-Time StreamingWatch deep agents work in real time. Every step, every delegation, every reflection — streamed as events.4 code cells3 with outputAPI checked
- 17Deep agentsDeep Agents: Real-World ExamplesFull-power deep agents with web search, built-in tools, MCP servers, streaming, and channels. Everything a production agent system needs.6 code cellsAPI checked
- 18Deep agentsDeep Agents with MemoryDeep agents that remember across runs. Conversation history, learned facts, and tracked entities persist so the agent builds on previous work.5 code cells3 with outputAPI checked
- 19Deep agentsSupervisor with Real ToolsWorkers that actually search the web, run code, and fetch pages. Three ways to give workers tools: with_builtins (all tools), specific tools, or Agent.with_builtins.4 code cellsAPI checked
- 20Deep agentsGoalAgent with Real ToolsGoalAgent that uses web search, code execution, and URL fetching to accomplish goals autonomously. The agent searches for real data, verifies it, and tracks success criteria.4 code cellsAPI checked
- 21Deep agentsReflectiveAgent with Web ResearchReflectiveAgent that searches the web to verify facts, then reflects on quality, and revises until the threshold is met. The agent doesn't just write — it fact-checks itself.3 code cellsAPI checked
- 22RAGSuper RAG BasicsThis notebook covers the standalone RAG subsystem at shipit_agent.rag — no agents yet, just indexing and retrieval. Topics:12 code cells11 with outputAPI checked
- 23RAGRAG + AgentPlug the Super RAG subsystem into a regular shipit_agent.Agent using one constructor parameter — rag=. The agent gains three new tools (rag_search, rag_fetch_chunk, rag_list_sources) and the system prompt is augmented…9 code cells7 with outputAPI checked
- 24RAGRAG + Deep AgentsAll five deep-agent patterns in shipit_agent.deep accept the same rag= parameter as Agent. They forward it to every inner Agent they build, so every step of a multi-step agentic workflow has access to the knowledge base.10 code cells9 with outputAPI checked
- 25Deep agentsDeepAgent + live chatThis notebook tours all of it.12 code cells11 with outputAPI checked
- 26Deep agentsAgent + Memory (OpenAI-style)How to give a shipit_agent.Agent and a DeepAgent long-term memory — the kind that remembers facts across sessions, not just within one chat. Three ingredients:10 code cells9 with outputAPI checked
- 27SkillsSkills Catalog, Tool Bundles & Power FeaturesFull tour: catalog browse, tool bundles, memory, streaming, chat sessions, and real-world project building with skills.21 code cells7 with outputAPI checked
- 28SkillsCreate And Use Custom SkillsUse this notebook as a guide for authoring a new skill, saving it into a local catalog, and attaching that skill to both Agent and DeepAgent.10 code cellsAPI checked
- 29SkillsDeepAgent With Skills, Memory & VerificationThis notebook demonstrates the full power of DeepAgent combined with skills:11 code cellsAPI checked
- 30SkillsBuild a Full Project With Skills & ToolsThis notebook uses the agent to actually create a real project from scratch. The full-stack-developer skill auto-attaches 13 tools including write_file, edit_file, bash, run_code, workspace_files, and plan_task.11 code cells11 with outputAPI checked
- 31Teams, costs and notificationsLoad Trace33 code cells17 with outputAPI checked
- 32Teams, costs and notificationsPrebuilt AgentsLoad production-ready agent personas in one line. 40 agents across 8 categories — Architecture, Code Quality, Security, DevOps, Testing, Planning, Research, and Content.14 code cells13 with outputAPI checked
- 33Teams, costs and notificationsShipCrew: Multi-Agent OrchestrationOpenClaw-style multi-agent crews where specialized agents collaborate on tasks autonomously. DAG-based task dependencies, parallel execution, and template variable resolution.18 code cells7 with outputAPI checked
- 34Teams, costs and notificationsNotification Hub: Slack, Discord & TelegramPush agent status updates to wherever your team lives. Get notified when agents start, complete, fail, or exceed budgets.13 code cells12 with outputAPI checked
- 35Teams, costs and notificationsCost Tracking & BudgetsKnow exactly what your agents cost. Set budget limits. Get alerts before overspending.15 code cellsAPI checked
- 36SkillsSkill Power Tools And MetadataThis notebook shows how skills now do two jobs at runtime:6 code cellsAPI checked
- 37AutopilotAutopilot quickstartAutopilot is the long-running, goal-driven, budget-gated runner. It wraps GoalAgent + PersistentAgent with:5 code cellsAPI checked
- 38AutopilotAutopilot live streamingEvery long-running run should show signs of life. This notebook demos Autopilot.stream() — a generator that yields events in real time, and live_renderer.render_stream() — a pretty TUI consumer. Drop-in components for…6 code cellsAPI checked
- 39AutopilotPersistence, resume, and the scheduler daemonAutopilot checkpoints every iteration so a crash is cheap. The SchedulerDaemon takes that further — a persistent goal queue on disk, drained tick-by-tick until SIGINT. This is the piece that lets shipit_agent run for…6 code cellsAPI checked
- 40Specialist agentsDeveloper · Debugger · ResearcherThree of the 47 specialists that ship with shipit_agent. Each is a fully-formed prompt + model + tool preset. Use them alone, or hand any of them to an Autopilot for long-running operation.6 code cellsAPI checked
- 41Specialist agentsDesign · PM · Sales · CS · MarketingThe non-engineering side of the specialist roster. Each persona is prompted for domain-correct outputs — a design reviewer leads with user tasks, a marketing writer cuts adjectives, a CS manager opens with the…8 code cellsAPI checked
- 42Specialist agentsPower toolsThree new tools from this release — each pairs naturally with one or more of the specialists from notebook 41.8 code cellsAPI checked
- 43Specialist agentsFan-out · Critic · ArtifactsThree features that turn Autopilot from a solid long-runner into a desktop-grade coordinator:9 code cellsAPI checked
- 44Specialist agentsThe Complete TourOne notebook that exercises every new feature this release ships with. Use it as a runnable reference when showing the library to someone for the first time — it starts with the LLM bootstrap and ends with a multi-agent…17 code cellsAPI checked
- 45Specialist agentsv1.0.6 extras · cost router · async ask_user · vision · sandbox · specialists-as-developersThe second half of v1.0.6 — five additional primitives that compose with the Autopilot core from the first half (notebooks 37–44). Each one composes with Autopilot, the critic, fan-out, and artifacts — no special wiring.16 code cellsAPI checked
- 46Specialist agentsDashboard Render Tool + LiteLLMThis notebook shows three things end-to-end:7 code cells7 with outputAPI checked
- 47Specialist agentsPM — Weekly PR Digest Across 3 ReposEnd-to-end workflow:8 code cellsAPI checked
- 48Specialist agentsDesigner — Figma Review WorkflowEnd-to-end workflow:6 code cellsAPI checked
- 49Specialist agentsSales — Lead Enrichment → Outreach → Salesforce ActivityEnd-to-end workflow:8 code cellsAPI checked
- 50Specialist agentsManager — Weekly KPI Dashboard From SheetsEnd-to-end workflow:6 code cellsAPI checked
- 51Specialist agentsCustomer Support — Zendesk Triage + VisionEnd-to-end workflow:7 code cellsAPI checked
- 52Specialist agentsData Analyst — SQL Question to DashboardUnlike the other persona notebooks, this one does not stub the database — we spin up an in-memory SQLite engine, seed it with a small sales table, then drive SQLTool against it for real. That gives you one notebook…6 code cellsAPI checked
- 53Specialist agentsFinance Analyst — Stripe + Contract PDF → Cash-Flow One-PagerWorkflow:7 code cellsAPI checked
- 54Reliability and controlStructured Output with Auto-Retry (v1.0.8)Get reliably-typed JSON / Pydantic output from any LLM, with three guarantees:8 code cellsAPI checked
- 55Reliability and controlVerifier Network — Process Supervision (v1.0.8)Catch hallucinated tool calls and stalled agents before they waste tokens.5 code cellsAPI checked
- 56Reliability and controlEpisodic Memory Consolidation (v1.0.8)Turn long conversations into durable facts. Apply forgetting curves so old facts decay. Promote frequently-retrieved facts to a 'core memory' set always visible in the system prompt.5 code cellsAPI checked
- 57Reliability and controlTime-Travel Replay (v1.0.8)Load any saved agent trace, fork from any event, edit the prompt, resume on a fresh agent. Side-by-side diff between two runs.8 code cellsAPI checked
- 58Reliability and controlComputerUseAgent — Browser Automation with Live StreamingThe screenshot → reason → act loop, now fully observable:19 code cells8 with outputAPI checked
- 59Reliability and controlBrowserAgentTool — ComputerUseAgent inside the main Agent (v1.0.8)The killer pattern: your main Agent plans, and when it needs to drive a browser it delegates to a sub-ComputerUseAgent via a single browser_use tool call.6 code cells4 with outputAPI checked
- 60Reliability and controlMultimodal Chat — inline media references in user prompts (v1.0.9)Users paste images, audio, video, or PDFs anywhere in a prompt:16 code cellsAPI checked
- 61Reliability and controlControl plane — permissions, plan mode & blocking hooksEvery cell runs offline — we drive the agent with a tiny scripted LLM, so no API keys are needed.11 code cellsAPI checked
- 62Provider API featuresPrompt caching — 10× cheaper repeated callsAnthropic, Bedrock-Anthropic and Vertex-Anthropic support prompt caching: the stable prefix of a request — the system prompt and the tool definitions — is cached on the provider side. On a cache hit, that prefix is…7 code cellsAPI checked
- 63Provider API featuresCross-session memory toolEvery path is confined to root_dir (default .shipit_workspace/memories); paths that escape the sandbox are rejected. This notebook drives the tool directly to demonstrate each command offline, then attaches it to an…10 code cellsAPI checked
- 64Provider API featuresServer-side tools (zero-infra)Anthropic hosts a set of server-side tools that run inside Anthropic's own sandbox: the model invokes them, Anthropic executes them, and the results come back inline in the response. There is no client-side tool loop…7 code cells6 with outputAPI checked
- 65Provider API featuresCitations + Batch APITwo v1.0.12 features in one notebook:9 code cells8 with outputAPI checked
- 66Provider API featuresInterleaved thinking & context editingTwo Anthropic beta passthroughs exposed on AnthropicChatLLM:7 code cells6 with outputAPI checked
- 67Workspace, live UI and observabilityThe SHIPIT WorkspacePoint an agent at a repository and it just works. The SHIPIT Workspace (shipit_agent.workspace) is a set of zero-code, file-based conventions you check into a project:10 code cells7 with outputAPI checked
- 68Workspace, live UI and observabilityLive task tracking with TodoToolThis notebook shows:6 code cells5 with outputAPI checked
- 69Workspace, live UI and observabilityGemma on Amazon Bedrock — the super agent, streaming, clean logsGoogle's Gemma 4 does native function calling on Bedrock, and shipit routes it transparently through the one class you already know — BedrockChatLLM. This notebook goes end to end:13 code cells11 with outputAPI checked
- 70Workspace, live UI and observabilityThe Super Agent: every sector, clean logs, real deliverablesThis notebook runs fully offline — a scripted LLM stands in so you can execute every cell with no API keys.9 code cells9 with outputAPI checked
- 71Workspace, live UI and observabilityThe Full Test Drive: every capability, one by oneTest the whole shipit super agent in depth, section by section:15 code cells15 with outputAPI checked
- 72Workspace, live UI and observabilityGuardrails & Deep Research: the protection + research layerTwo new capabilities, tested in depth:9 code cells9 with outputAPI checked
- 73Workspace, live UI and observabilityThe shipit CLI & Power Tools Test DriveEverything from the CLI round, exercised offline:7 code cells7 with outputAPI checked
- 74Workspace, live UI and observabilityLive streaming — every v1.1.0 feature against Gemma 4 on BedrockWatch a real agent work: human-verb transcript rows, deferred approvals, lockdown, delegated sub-agents, connections, and SSE frames — all against bedrock-mantle/google.gemma-4-26b-a4b.31 code cells31 with outputAPI checked
- 75Workspace, live UI and observabilityLive streaming, the live UI, and sub-agents — end to endA real agent, a real model (bedrock-mantle/google.gemma-4-26b-a4b), real tools against real files on disk. Nothing here is mocked, and nothing is a picture of what the output would look like — every panel below is drawn…40 code cells39 with outputAPI checked
- 76Workspace, live UI and observabilityApps, analysis, and the whole UI — end to endThe reference product's screenshot is a chat on the left and a thing on the right: a revenue dashboard, a spreadsheet, a workflow. The chat is where you ask; the thing is what you keep.23 code cells22 with outputAPI checked
- 77Workspace, live UI and observabilitySuper Agent - high-depth live validationThis notebook validates the main Agent as a long-running coding agent: optimized project setup, complete capability discovery, skills, MCP provenance, permission gates, durable sessions, cache accounting, multimodal…7 code cellsAPI checked
- 78Workspace, live UI and observabilityUse this checkout, not the installed packageRun this before anything else, and re-run it after a kernel restart.23 code cells20 with outputAPI checked
- 79Workspace, live UI and observabilityObservability with shipit-watcherOne agent run, reported five ways: traces, the agent graph, scores, datasets and the local ledger.10 code cells10 with outputAPI checked