# Gwenflow

> Documentation of Gwenflow, the open-source Python framework by Gwenlake for building AI agents and multi-agent workflows: LLM providers, tools, memory, flows, RAG and durable execution.

## Docs

- [Introduction](/gwenflow/index): Build AI agents and multi-agent workflows in Python: any LLM provider, tools, memory, flows, RAG and durable execution.
- [Quickstart](/gwenflow/quickstart): Install the library, call a model and run your first agent.
- [Installation](/gwenflow/installation): Python versions, extras, and the optional packages each feature needs.
- [Command line](/gwenflow/cli): Ask an agent, run a flow file or let AutoFlow plan a workflow from a terminal.
- [Examples](/gwenflow/examples): Runnable scripts in the repository, by theme.
- [Overview](/gwenflow/concepts/overview): The building blocks of Gwenflow and how a run moves between them.
- [Messages and responses](/gwenflow/concepts/messages): The provider-neutral types that go in and come out of models and agents.
- [Models](/gwenflow/guides/models): One API over OpenAI, Anthropic, Google, Mistral, Azure, DeepSeek, Ollama, Gwenlake and any OpenAI-compatible server.
- [Reasoning](/gwenflow/guides/reasoning): Read the reasoning of thinking models the same way, whatever the provider.
- [Multi-modal input](/gwenflow/guides/multimodal): Send images, audio and PDFs alongside text, with the same message for every provider.
- [Agents](/gwenflow/guides/agents): The agentic loop: instructions, tools, history, limits and usage.
- [Tools](/gwenflow/guides/tools): Turn Python functions into tools, write tool classes, and control retries and output size.
- [Built-in tools](/gwenflow/guides/built-in-tools): Search, research, finance, files, code execution and retrieval tools that ship with the library.
- [Structured output](/gwenflow/guides/structured-output): Get a validated Pydantic object from an agent, with or without tools.
- [Streaming](/gwenflow/guides/streaming): Consume an agent's answer, reasoning and tool calls as they happen.
- [Tool approvals](/gwenflow/guides/approvals): Pause a run on sensitive tools until a human approves or denies the call.
- [Teams](/gwenflow/guides/teams): Give an agent a team of specialists it delegates to through handoff tools.
- [Sub-agents](/gwenflow/guides/subagents): Let an agent launch fresh agents on demand for self-contained tasks, in parallel.
- [Coding agent](/gwenflow/guides/coding-agent): An agent preset that reads, edits, runs and verifies code inside one directory.
- [Skills](/gwenflow/guides/skills): Instructions, scripts and resources an agent loads only when a task calls for them.
- [MCP](/gwenflow/guides/mcp): Use the tools of Model Context Protocol servers, over stdio or HTTP.
- [Context management](/gwenflow/guides/context): How an agent keeps its history within the model's context window, and how to tune it.
- [Long-term memory](/gwenflow/guides/long-term-memory): Markdown files an agent reads and edits, so it remembers from one conversation to the next.
- [Flows](/gwenflow/guides/flows): Chain agents and data steps in a DAG, defined in YAML, JSON or Python, with independent branches in parallel.
- [Node types](/gwenflow/guides/flow-nodes): The built-in node types of a flow: agents, HTTP, readers, writers, databases, search and Python.
- [AutoFlow](/gwenflow/guides/autoflow): Let a model plan a team of agents for a query, then review, save or run the flow.
- [Retrieval (RAG)](/gwenflow/guides/rag): Read documents, split them, index them in a vector store and ground the agent's answers in them.
- [Durable execution with Temporal](/gwenflow/guides/durability): Run agents and flows on Temporal so a run survives crashes, deployments and restarts, and resumes where it stopped.
- [Telemetry](/gwenflow/guides/telemetry): Trace every agent run, LLM call, tool call and flow node with OpenTelemetry, into any OTLP backend.
- [Errors and logging](/gwenflow/guides/errors): What fails a run, what the model gets to recover from, and how to see what happens.
- [Changelog](/gwenflow/changelog): What changed in the gwenflow library, release by release.
- [Contributing](/gwenflow/contributing): Set up a development environment, run the tests, lint, and audit the dependencies.