Provara Docs
Self-hostable LLM operations platform — adaptive routing, regression detection, cost migration, and FinOps-grade spend intelligence.
What Provara is
Provara is an LLM operations platform. It sits between your application and the upstream LLM providers (OpenAI, Anthropic, Google, Mistral, xAI, Z.ai, Ollama, or any OpenAI-compatible endpoint) and delivers:
- Adaptive routing — picks the best model per request from real judge scores and user feedback, not a static pre-trained classifier
- Silent-regression detection — replays your top historical prompts against current models and flags when quality drops
- Auto cost migration — nightly job moves routing cells to cheaper models when quality parity holds, with one-click rollback
- Spend intelligence — FinOps-grade dashboard: per-user attribution, MTD + anomaly detection, savings recommendations, budgets with hard-stop
- Audit logs — compliance-ready, tier-gated retention, SIEM-friendly pull API
- Guardrails — PII detection, content policies, custom regex, blocking/redaction
- A/B testing — manual plus auto-generated experiments on tied cells
Everything speaks the OpenAI chat-completions format. Any SDK that points at an OpenAI-compatible URL works.
Start here
- Quick start — Docker or local dev in 5 minutes
- Architecture — the 10,000-foot picture
- Self-host vs Cloud — pick your path
Where to look when things break
- Incident response — top-level "gateway is broken" playbook
- Master-key rotation
- Backup & restore
License
Provara is licensed under the Business Source License 1.1 (BSL). Non-production use is free; commercial production use by larger organizations requires a license. Each release converts to Apache 2.0 four years after its release date. See CONTRIBUTING.md for the full summary.