FAL-OS is gourmet-grade software for frontier AI practitioners. It is an alignment-first operating system for crafting multi-agent systems with verifiable guardrails, observability, and synthetic data tooling.
- Curated craftsmanship – type-safe APIs, delightful CLI, rigorous tests, and reproducible scaffolds.
- Alignment by default – policy DSL, automated red teaming, and audit-ready traces built into every execution path.
- Model & infra agnostic – register open checkpoints, hosted APIs, or local runtimes through a unified adapter palette.
- Agent graph studio – declarative multi-agent workflows with memory, branching policies, and governance hooks.
- Complete observability – structured logs, metrics, experiment registry, and OTLP-compatible tracing exporters.
python3 -m venv .venv
source .venv/bin/activate
pip install -e .[dev]
# Diagnostics & inventory
falos doctor
falos models list
# Run the default agent graph
falos agents run --prompt "Summarise the latest alignment research"
# Launch the REST API
falos api --host 0.0.0.0 --port 8000With the API running, explore:
GET /api/v1/system/health– service heartbeatGET /api/v1/models– registered model profilesPOST /api/v1/agentgraphs/run– execute a graph ({"prompt": "..."})POST /api/v1/synthetic/run– trigger synthetic campaign (list of prompts)
| Module | Purpose |
|---|---|
falos.router |
Adapter registry, routing policies, benchmarking hooks |
falos.agentgraph |
YAML graph loader & async orchestrator |
falos.governance |
Policy specs, enforcement engine, verdict reporting |
falos.synth |
Synthetic campaign runner, self-play pipelines |
falos.observability |
Trace ledger, logging & OTLP exporters |
falos.eval |
Lightweight benchmark harness |
See docs/vision.md for the deep dive, docs/architecture.md for component
interaction, and docs/api.md (new) for REST contract details.
pytest # unit + integration suite using mock adapters
ruff check src tests
mypy src testsPre-commit hooks (pre-commit install) enforce formatting and type hygiene.
frontier_alignment_lab_os/
├── docs/ # design notes, architecture, roadmap
├── examples/ # agent graphs, policies, sample manifests
├── src/falos/ # core application code
├── tests/ # automated test suites
├── tools/ # developer scripts & notes
└── pyproject.toml
Read CONTRIBUTING.md for guidelines. We welcome adapters, policy packs, eval
suites, and observability integrations from the alignment community.
Apache 2.0 © 2025 Frontier Alignment Lab.