inspect-ai
Framework for large language model evaluations
Description
[<img width="295" src="https://inspect.aisi.org.uk/images/aisi-logo.svg" />](https://aisi.gov.uk/) Welcome to Inspect, a framework for large language model evaluations created by the [UK AI Security Institute](https://aisi.gov.uk/). Inspect provides many built-in components, including facilities for prompt engineering, tool usage, multi-turn dialog, and model graded evaluations. Extensions to Inspect (e.g. to support new elicitation and scoring techniques) can be provided by other Python packages. To get started with Inspect, please see the documentation at <https://inspect.aisi.org.uk/>. Inspect also includes a collection of over 200 pre-built evaluations ready to run on any model (learn more at <https://inspect.aisi.org.uk/evals/>). *** To work on development of Inspect, clone the repository and install with the `-e` flag and `[dev]` optional dependencies: ```bash git clone https://github.com/UKGovernmentBEIS/inspect_ai.git cd inspect_ai pip install -e ".[dev]" ``` Optionally install pre-commit hooks via ```bash make hooks ``` Run linting, formatting, and tests via ```bash make check make test ``` If you use VS Code, you should be sure to have installed the recommended extensions (Python, Ruff, and MyPy). Note that you'll be prompted to install these when you open the project in VS Code. ### Frontend development (TypeScript) The web UI lives in a git submodule at `src/inspect_ai/_view/ts-mono/`. **These steps are only needed if you plan to work on the TypeScript/React frontend** — Python-only contributors can skip this entirely. Initialize the submodule and install dependencies — see the [one-time setup guide](src/inspect_ai/_view/ts-mono/docs/submodule-guide.md#one-time-setup). ### Documentation To work on the Inspect documentation, install the optional `[doc]` dependencies with the `-e` flag and build the docs: ``` pip install -e ".[doc]" cd docs quarto render # or 'quarto preview' ``` If you intend to work on the docs iteratively, you'll want to install the Quarto extension in VS Code.
Release History
| Version | Changes | Urgency | Date |
|---|---|---|---|
| 0.3.284 | ## [0.3.284](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.283...0.3.284) (2026-10-25) ### Features * **moonshot:** support Kimi K3.1 Pro and K3.1 Pro Highspeed ([#6072](https://github.com/UKGovernmentBEIS/inspect_ai/issues/6072)) ([eade928](https://github.com/UKGovernmentBEIS/inspect_ai/commit/eade928ea4119e0f0eddfe58617d9993ba9d710a)) * **sentinel:** step.agent_messages is the running agent's own message list at every stage ([#5986](https://github.com/UKGovernmentBEIS/inspect_a | High | 10/25/2026 |
| 0.3.283 | ## [0.3.283](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.282...0.3.283) (2026-10-21) ### Breaking Changes * dataset loaders now read a top-level `checkpoint` field as the sample's checkpoint configuration and raise when it isn't a valid one. Datasets that use `checkpoint` for unrelated data should pass `FieldSpec(checkpoint=None)` to skip the field, or `FieldSpec(checkpoint="<column>")` to read checkpoint settings from another column. Dataset records also can't set the checkpoi | High | 10/21/2026 |
| 0.3.282 | ## [0.3.282](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.281...0.3.282) (2026-10-19) ### Breaking Changes * tool state scopes so agents running concurrently in one sample keep separate tool state ([#5861](https://github.com/UKGovernmentBEIS/inspect_ai/issues/5861)) ### Features * **sharding:** conditional S3 writes and a local lock helper ([#5775](https://github.com/UKGovernmentBEIS/inspect_ai/issues/5775)) ([5384f73](https://github.com/UKGovernmentBEIS/inspect_ai/commit/5384 | High | 10/19/2026 |
| 0.3.281 | ## [0.3.281](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.280...0.3.281) (2026-10-15) ### Breaking Changes * require Python 3.11 and test on Python 3.14 ([#5862](https://github.com/UKGovernmentBEIS/inspect_ai/issues/5862)) ### Features * add swarm_controller and swarm_channel registry types ([#5860](https://github.com/UKGovernmentBEIS/inspect_ai/issues/5860)) ([a11afc9](https://github.com/UKGovernmentBEIS/inspect_ai/commit/a11afc94322a431782f57c0f48a3d6b7666413c3)) * **agent-b | High | 10/15/2026 |
| 0.3.280 | ## [0.3.280](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.279...0.3.280) (2026-10-14) ### Features * **agent-bridge:** filters can now tell which agent a bridged model request belongs to ([#4791](https://github.com/UKGovernmentBEIS/inspect_ai/issues/4791)) ([1b78969](https://github.com/UKGovernmentBEIS/inspect_ai/commit/1b78969d6350bd8ce84bba84a92cbaddf8b175d6)) * **google:** support Gemini 4 Argon ([#5843](https://github.com/UKGovernmentBEIS/inspect_ai/issues/5843)) ([76e69f9]( | High | 10/14/2026 |
| 0.3.279 | ## [0.3.279](https://github.com/UKGovernmentBEIS/inspect_ai/compare/0.3.278...0.3.279) (2026-10-12) ### Bug Fixes * Internal improvements. | High | 10/12/2026 |
| 0.3.278 | - Tasks and samples can carry a `description` of what they ask of the agent, recorded in the eval log and in `evals_df()` and `samples_df()`. - Tool review now also reviews `handoff()` calls, as approval does. - Compaction: Fixed compaction triggering early and reporting inflated token counts on OpenAI reasoning models, which counted replayed reasoning twice. - Compaction: Removed `ModelAPI.apply_redacted_reasoning_tokens_to_input()` and stopped adding `redacted_reasoning_tokens` to assistant me | High | 10/9/2026 |
| main@2026-10-09 | Latest activity on main branch | High | 10/9/2026 |
| 0.3.209 | Imported from PyPI (0.3.209) | Low | 4/21/2026 |
| 0.3.208 | Tag 0.3.208 | High | 4/19/2026 |
| 0.3.207 | Tag 0.3.207 | High | 4/16/2026 |
| 0.3.206 | Tag 0.3.206 | High | 4/15/2026 |
| 0.3.205 | Tag 0.3.205 | Medium | 4/4/2026 |
| 0.3.204 | Tag 0.3.204 | Medium | 4/2/2026 |
| 0.3.203 | Tag 0.3.203 | Medium | 4/1/2026 |
| 0.3.202 | Tag 0.3.202 | Medium | 3/31/2026 |
| 0.3.201 | Tag 0.3.201 | Medium | 3/25/2026 |
| 0.3.200 | Tag 0.3.200 | Low | 3/20/2026 |
| 0.3.199 | Tag 0.3.199 | Low | 3/17/2026 |
| 0.3.197 | Tag 0.3.197 | Low | 3/17/2026 |
| 0.3.196 | Tag 0.3.196 | Low | 3/16/2026 |
| 0.3.195 | Tag 0.3.195 | Low | 3/15/2026 |
| 0.3.193 | Tag 0.3.193 | Low | 3/13/2026 |
| release/2025-11-28 | Tag release/2025-11-28 | Low | 11/28/2025 |
| inspect-tool-support-1.2.0 | Tag inspect-tool-support-1.2.0 | Low | 8/11/2025 |
| inspect-tool-support-1.1.2 | Tag inspect-tool-support-1.1.2 | Low | 7/9/2025 |
| inspect-tool-support-1.1.0 | Tag inspect-tool-support-1.1.0 | Low | 5/2/2025 |
| STASH_ANYIO_PASS1 | Tag STASH_ANYIO_PASS1 | Low | 4/10/2025 |
