FlightDeck¶
Ship AI agents safely — release diffs, runtime evidence, and policy gates.
FlightDeck is a local-first CLI + SQLite + optional web UI that versions your agent builds, collects runtime evidence (cost, latency, errors), diffs a baseline against a candidate, and blocks promote until policy passes. Apache-2.0.
Try it — 30 seconds, no account needed¶
pip install flightdeck-ai
flightdeck demo
flightdeck demo runs the full register → ingest → diff → promote loop in a temporary workspace. Nothing leaves your machine.
How it works¶
- Register an immutable release snapshot (
release.yaml+ checksum). - Ingest runtime evidence from your agent (cost, latency, errors, confidence).
- Diff baseline vs candidate — cost per run, latency delta, error rate, pricing catalog.
- Promote only when policy passes; optional human approval before the ledger moves.
The same contract works from the CLI, the HTTP API (POST /v1/promote), and the bundled web UI (flightdeck serve).
Where to start¶
| I want to… | Go here |
|---|---|
| Try it in 30 seconds | pip install flightdeck-ai && flightdeck demo |
| Wire it to my real agent | Getting started |
| See all CLI commands | CLI reference |
| Use the HTTP API | HTTP API |
| Understand policy gates | Operations & policy |
Self-host flightdeck serve |
HTTP API — starting the server |
| Deploy for a team in production | Getting started → Production checklist |
| Something broke | Troubleshooting |
Quick reference¶
| Topic | Doc |
|---|---|
| Commands, flags, exit codes | CLI reference |
flightdeck serve JSON API |
HTTP API |
| Diff, promote, rollback, SQLite | Operations & policy |
release.yaml, workspace config |
Release artifact |
| Optional pricing catalog YAML | Pricing catalog |
flightdeck Python client |
Python SDK |
| Experimental adoption hooks | SDK integrations |
| Shipped web UI vs roadmap | Web UI · UI roadmap |
| Common failures | Troubleshooting |
Full examples and CI templates: github.com/flightdeckdev/flightdeck/examples
Who should use this?¶
- ML / platform engineering teams shipping LLM agents to production who want a policy-gated promote path — not just a dashboard.
- Regulated or compliance-sensitive teams (fintech, healthcare) where data residency and audit trails matter. Local-first by default; self-host
flightdeck serveto keep data on-prem. - Engineers who want to answer "is this candidate safe to ship?" with numbers and policy, not gut feel.
Install options¶
User (recommended):
pip install flightdeck-ai
Optional extras: flightdeck-ai[openai], flightdeck-ai[anthropic], flightdeck-ai[postgres], flightdeck-ai[telemetry]
Contributor (uv):
git clone https://github.com/flightdeckdev/flightdeck
cd flightdeck
uv sync --extra dev
uv run flightdeck --help
See DEVELOPMENT.md for web bundle, schema generation, and CI parity.
Ask AI¶
Use the floating Ask AI button (bottom-right) to open Perplexity with this docs site and the GitHub repo as context. No FlightDeck servers are involved.