What you get

• A LangGraph service, ready to run: a streaming chat API, threads, an A2A endpoint, answers in a JSON shape you declare when a program reads them, and shared-key, OIDC/JWT or custom authentication on every route.
• Outbound calls under a policy: api-policy.yaml declares the APIs tools may call, the agent refuses anything else, lint checks it in CI, and chosen calls wait for a human approval.
• Agents that ask other agents: peer add declares the agents one asks over A2A, each agent knows the user and the agent in between, a person's approvals stay with that person, and system checks, wires and deploys several projects as one.
• An eval gate CI can enforce: deterministic checks and model judges, with an exit code that is the gate and a deterministic fake model for keyless runs.
• Deploys to any cluster: a hardened Helm chart, deploy straight from your machine or through GitHub Actions or Argo CD, and secrets for the app's Secret.
• Works with your coding agent: skills teach Claude Code, Codex, Gemini CLI, Cursor, Antigravity and others the same lifecycle.

Nothing is specific to one domain or one consumer: a project picks its auth policy, declares its APIs and configures the rest through environment variables and chart values.

Install

You need Python 3.12 or 3.13 and uv. Deploying also needs helm, kubectl, git and a docker that builds with BuildKit.

uv tool install graph-agents-cli
graph-agents-cli setup      # optional: install the skills into your coding agents

pipx install graph-agents-cli or pip install graph-agents-cli (into a virtual environment) work too. To install a release tag from GitHub instead:

uv tool install git+https://github.com/ss7172/graph-agents-cli@v0.3.1

More options (extras, mirrors, disconnected installs): Installation & setup.

Quick start

Run a new agent locally. The fake model needs no key; for a real one, leave out the export line and login --write-env asks for OPENAI_API_KEY.

graph-agents-cli create my-agent && cd my-agent
cp .env.example .env
export MODEL_PROVIDER=fake            # the keyless, deterministic test model
graph-agents-cli login --write-env    # checks the setup, generates API_KEY
graph-agents-cli install
graph-agents-cli run "What's the weather in San Francisco?"
graph-agents-cli eval run             # the exit code is the gate

Or ask your coding agent to "use graph-agents-cli to build ..." once setup has installed the skills. The Quickstart walks through each step and switches to a real provider.

Documentation

The documentation site is built with MkDocs Material from website/.

To preview the site from a checkout, for example while editing it:

uv run --group docs mkdocs serve -f website/mkdocs.yml    # http://127.0.0.1:8200

On the command line, graph-agents-cli <command> --help prints the same reference as the site's CLI page.

Status

Version 0.3.1, alpha (release notes). Interfaces may still change between minor versions; the changelog lists every breaking change with its migration steps. Released on PyPI and as tags on GitHub.

• Known issues: parked issues, each with its impact and workaround.
• Changelog: every release and its migration steps.
• Contributing: development setup, tests, templates, releases and the upstream-sync process.

License

Apache-2.0: see LICENSE. graph-agents-cli is a fork of google-agents-cli with the Google Cloud specific parts removed; NOTICE lists the attribution and the modifications.