agentic-graph-composer

Agentic Graph Composer

A CLI- and canvas-first builder for AI agents on LangGraph. Define an agent as a declarative YAML schema; Agentic Graph Composer compiles it into a real LangGraph StateGraph - not a reimplementation of LangGraph, a thin, typed layer on top of it.

Install

pip install agentic-graph-composer

This installs the agc command.

Quickstart

mkdir my-agent && cd my-agent
agc init                       # scaffolds a working schema.yaml + starter files
cp .env.example .env                  # then fill in your provider's API key
agc validate schema.yaml
agc run schema.yaml "What does this project do?"

agc doctor schema.yaml checks your environment (API keys, optional extras, checkpointer config) against what a schema actually needs, before you hit a failure mid-run.

The canvas (visual editor)

A visual editor for the graph a schema compiles to ships bundled in - no separate install, no Node.js needed:

agc canvas schema.yaml

This prints a URL - open it in a browser to see and edit the graph, backed by the same schema/compiler library the CLI uses. See canvas/README.md for the view-only mode and running the canvas from source.

What a schema looks like

schema_version: 1
nodes:
  - id: assistant
    llm:
      provider: anthropic
      model: claude-sonnet-5
      system: "You are a helpful assistant."
    tools:
      - tools:search_docs

A schema with exactly one node and no edges section is implicitly wired START -> node -> END. Multi-node graphs add explicit edges (direct or conditional), a handler node for custom-code logic outside the schema, and an opt-in checkpointer block for resumable runs. See docs/GETTING_STARTED.md for the full setup guide - extras, provider API keys, Postgres checkpointing, observability. See docs/README.md for the complete design-doc suite.

Optional extras

pip install "agentic-graph-composer[examples]"   # langchain-anthropic, langchain-openai - to run examples/
pip install "agentic-graph-composer[postgres]"   # Postgres-backed checkpointing

Learn more

• docs/GETTING_STARTED.md - the practical setup guide
• docs/README.md - architecture, ADRs, requirements, roadmap
• CONTRIBUTING.md - developing Agentic Graph Composer itself
• CHANGELOG.md

License

MIT