Inspect AI
Government-built open-source framework for rigorous LLM and agent evaluations, popular for safety benchmarks and sandboxed agentic tasks.
| Category | Evals & Testing |
| Open source | Yes |
| Self-hostable | Yes |
| Pricing model | free |
| Pricing notes | MIT OSS, free; no commercial tier |
| Framework integrations | openai-sdk, anthropic, ollama, huggingface |
| Funding / ownership | Built by the UK AI Security Institute (government) |
| GitHub stars | 2,194 |
| Maintenance | active (commit 2026-06-12) |
| License (GitHub) | MIT |
| Open issues | 212 |
| Maturity signal | 91/100 (Mature) — computed from public GitHub signals, see methodology |
Pricing/feature source: https://github.com/UKGovernmentBEIS/inspect_ai
Maturity signal is computed from public GitHub data only. How it is calculated.
Frequently asked questions
Is Inspect AI open source?
Yes. Licensed under MIT.
Can I self-host Inspect AI?
Yes — Inspect AI can run in your own infrastructure.
How much does Inspect AI cost?
Free. MIT OSS, free; no commercial tier