Skip to main content

API & CLI

The agentflow CLI (10xscale-agentflow-cli) provides every command you need to scaffold, run, test, evaluate, and containerize your agents.

Installation​

pip install 10xscale-agentflow-cli

Verify the installation:

agentflow --help

Commands​

CommandDescription
agentflow initInteractively scaffold a new agent project
agentflow apiStart the FastAPI development server
agentflow playStart the server and open the hosted playground
agentflow buildGenerate a Dockerfile (and optionally docker-compose.yml)
agentflow skillsInstall bundled coding-agent skills (Codex, Claude, GitHub)
agentflow testRun the project test suite via pytest
agentflow evalRun agent evaluations and generate HTML + JSON reports
agentflow versionPrint CLI and package version

Command summaries​

agentflow init​

Scaffolds a new agent project interactively. Prompts for agent name and setup type (Quick Start or Production). For production projects, also prompts for authentication and rate-limiting configuration.

agentflow init                  # scaffold in the current directory
agentflow init --path ./my-bot  # scaffold in a specific directory
agentflow init --force          # overwrite existing files

See Initialize a project for the full guide.

agentflow api​

Starts a Uvicorn-backed FastAPI server that loads your compiled graph from agentflow.json. Auto-reload is enabled by default.

agentflow api
agentflow api --host 0.0.0.0 --port 8000
agentflow api --no-reload       # disable file-watching (production)

Default host: 127.0.0.1. Default port: 8000.

See Run the API server for the full guide.

agentflow play​

Same as agentflow api but also opens the hosted playground in your default browser once the server is reachable.

agentflow play
agentflow play --port 8001

See Open the playground for the full guide.

agentflow build​

Generates a production Dockerfile. Optionally generates docker-compose.yml as well (omitting the CMD from the Dockerfile in that case).

agentflow build
agentflow build --docker-compose
agentflow build --python-version 3.12 --port 8080
agentflow build --force         # overwrite existing Dockerfile

Default Python version: 3.13. Default service name in docker-compose: agentflow-api.

See Generate Docker files for the full guide.

agentflow skills​

Installs bundled AgentFlow coding-agent skills into your project for Codex, Claude, or GitHub Copilot.

agentflow skills                       # interactive agent selection
agentflow skills --agent claude
agentflow skills --agent codex
agentflow skills --agent github
agentflow skills --all                 # install for every supported agent
agentflow skills --list                # list supported agents
agentflow skills --force               # overwrite existing installation

See Install skills for the full guide.

agentflow test​

Thin pytest wrapper. Reads optional defaults (path, coverage, coverage_threshold) from agentflow.json. Extra arguments after -- are forwarded to pytest verbatim.

agentflow test
agentflow test tests/unit
agentflow test --coverage
agentflow test --coverage --html       # open HTML coverage report
agentflow test -k "test_graph"         # keyword filter
agentflow test -- --tb=short           # forward flags to pytest

See Run tests for the full guide.

agentflow eval​

Discovers *_eval.py / eval_*.py files, collects all cases into a flat pool, runs them under a single async event loop, and writes timestamped HTML + JSON reports to eval_reports/.

agentflow eval
agentflow eval evals/weather_agents_eval.py
agentflow eval --parallel --max-concurrency 8
agentflow eval --threshold 0.9         # fail if pass rate < 90 %
agentflow eval --no-report             # console summary only
agentflow eval --open                  # open HTML report in browser

Default output directory: eval_reports/. Default max concurrency: 4.

See Run evaluations for the full guide.

agentflow version​

Prints the CLI version constant and the installed package version read from pyproject.toml.

agentflow version

Example output:

agentflow-cli CLI
  Version: 1.0.0
agentflow-cli Package
  Version: 0.3.2.8

Global flags​

Every command accepts --verbose / -v (detailed logging) and --quiet / -q (errors only). Pass -h or --help to any command for its full flag reference.

Last updated on