Weave Documentation
Weave is a Python framework for building composable AI agents. It's a universal interface standard so any compliant agent can plug into any other, regardless of LLM, language, or host. USB for AI agents.
Start here
| Guide | What it covers |
|---|---|
| Getting Started | Install, define your first agent, run a pipeline |
| Architecture | The agent anatomy, module map, and data flow (see also architecture.html) |
| Catalog | Capability-tag vocabulary, data types, and the manifest format |
Guides
| Guide | What it covers |
|---|---|
| Agents | The @agent decorator and BaseAgent; the execution lifecycle |
| LLM Providers | Swapping brains, the adapter contract, custom providers |
| Memory | Short-term buffer and long-term vector memory |
| Guardrails | pre / post / on_error hooks |
| Connections & Pipelines | The connection protocol, transforms, Router, Pipeline, Parallel, LocalRunner |
| Interop (LangChain/CrewAI) | Wrap a foreign-framework agent as a Weave agent, no rewrite needed |
Reference
| Reference | What it covers |
|---|---|
| Public API | Every export from weaveflow |
| CLI | weaveflow scaffold / validate / package |
| Errors | Every WEAVE_* error code and how to fix it |
Core philosophy
An LLM is a brain. An agent is the body. The better the body (its sensors, memory, actuators, and nervous system), the better the output of the brain. Weave defines the standard anatomy of that body so any LLM can inhabit any agent structure.
Visual architecture
Open architecture.html in a browser for an interactive, dependency-free walkthrough of the framework's anatomy, module layers, agent lifecycle, and the connection protocol.
Building this docs site (later)
The Markdown here renders directly on GitHub. To serve a full docs site:
pip install -e ".[docs]"
make docs # serve locally at http://127.0.0.1:8000
make docs-build # build static site into ./site