# UKGovernmentBEIS/inspect_ai

Inspect: A framework for large language model evaluations

Repository: https://github.com/UKGovernmentBEIS/inspect_ai
Canonical: https://ross.abutalabs.com/products/inspect_ai
Homepage: https://inspect.aisi.org.uk/
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-26T21:47:05+00:00

## Health v2 (maintenance only)
Score: 71/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 73
- inputs: {"age_days": 1023, "days_push": 7, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2640, forks 679 (observed 2026-08-28T04:07:06.340015+00:00)

## What it is
Inspect is an open-source Python framework from the UK AI Security Institute for running large language model evaluations, with composable building blocks for datasets, solvers, tools, scorers, and agents. It ships 200+ pre-built benchmark evaluations, a web-based Inspect View for visualizing results, a VS Code extension, and sandboxing support for running untrusted model code across Docker, Kubernetes, Modal, and other backends.

## Use cases
- run LLM benchmarks on any model provider
- evaluate coding agents like Claude Code or Codex CLI on agentic tasks
- build custom evaluations with prompt engineering, tool use, and multi-turn dialog
- score model outputs with model-graded and custom scorers
- sandbox untrusted model-generated code safely during evals
- monitor and visualize evaluation logs in a web UI
- test agent behavior, reasoning, and safety properties
- run MCP and custom tools inside evaluations

## When to choose
- you need a rigorous, reproducible evaluation harness for LLMs or agents
- you want to run established benchmarks without writing them yourself
- you need sandboxed execution of model-generated code
- you want extensible Python tooling with a mature ecosystem and active development

## When to avoid
- you only need simple one-off model API calls without evaluation scoring
- you need a production inference-serving framework rather than an eval harness
- your stack is not Python-based and you cannot add Python tooling

## Facets
- artifact type: framework
- maturity: active
- function: llm-inference, agent-framework, testing, benchmarking, mcp, chatbot, machine-learning, developer-tools, cli, data-visualization
- domain: large-language-models, artificial-intelligence, machine-learning, developer-tools, testing
- platform: python, cloud, cli, cross-platform
- tags: llm-evaluations, eval-framework, benchmark-suite, sandboxing, model-graded-evals, vscode-extension, inspect-view, agentic-evals, ai-safety, prompt-engineering, ai-agents, command-line, docker, kubernetes

## Member repositories
- UKGovernmentBEIS/inspect_ai (main) score 71

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:06.340015+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:19:05.258795+00:00, confidence not recorded.
  - readme: https://github.com/UKGovernmentBEIS/inspect_ai (fetched 2026-08-28T04:07:06.340015+00:00, sha 40050b0474aa)
  - homepage: https://inspect.aisi.org.uk/ (fetched 2026-08-29T10:02:09.371060+00:00, sha 8a14eb433269)
  - registry_pypi: https://pypi.org/pypi/inspect_ai/json (fetched 2026-08-29T10:02:09.381675+00:00, sha c9c7f93f71de)
  - site_page: https://inspect.aisi.org.uk/CHANGELOG.html (fetched 2026-08-29T10:02:09.374044+00:00, sha 6c022f2fa60a)
- Data as of 2026-08-30T08:39:29.467469+00:00.