Benchside
An AI workspace for doing science, maintained by Science Works. Benchside provides eight research roles, private project records, saved decisions and unresolved questions, and optional reviewed contributions to metascience research.
Install for a local agent
Requires Python 3.11+ on macOS or Linux. Python 3.11 and 3.14 are tested. Benchside supplies the skills and runtime; your agent supplies the model and execution tools. It works with capable agents using OpenAI, Anthropic, other providers or locally hosted models.
With uv:
uv tool install benchside
bench setup --agent codex
Or install with pip in a Python virtual environment:
python -m pip install benchside
bench setup --agent codex
Use --agent claude for Claude Code. For another agent, use
bench setup --agent generic --skills-dir /path/to/that/agents/skills after
identifying its actual skill directory. The agent can run these steps for you.
Then ask it to read the installed Benchside skill and complete guided setup:
choose private storage, select a project, and verify saving and restoring a
synthetic record. Installing the Python package alone does not complete that setup.
The package is named benchside; its command is bench. Upgrade with
uv tool upgrade benchside or python -m pip install --upgrade benchside, then
rerun the same bench setup command to refresh the installed skill. Keep private
records in their existing location. Every app sharing a notebook upgraded by
0.4.1 must use 0.4.1 or newer, including apps previously running 0.4.0.
Research roles and records
- Research supervisor and focused hypothesis reviewer
- Study-design consultant and methods troubleshooter
- Results discussant and reproducibility auditor
- Journal-club discussant and scientific peer reviewer
The agent selects a suitable role with the researcher. Decisions and research records remain in the chosen private store, with ODT/Markdown reading copies for local storage. Optional sharing is disabled initially; scientific findings are excluded from metascience submissions. Workflow problems require review and approval. Survey participation is agreed before answering.
ChatGPT and Claude websites
This Python package does not install a web-app plugin or grant storage/network permissions. For those apps, use the matching archive from GitHub Releases and follow the setup guide.
Source and full documentation · Report an issue · Contribute a research role
Benchside is research software under active development, licensed under Apache 2.0.
1 maintainer
Metadata
1 maintainer
Release files for benchside 0.4.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| benchside-0.4.1.tar.gz | 216.3 kB | With provenance | Details for benchside-0.4.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| benchside-0.4.1-py3-none-any.whl | Python 3 | none | any | Details for benchside-0.4.1-py3-none-any.whl |
Total release size: 466.3 kB
Release files / benchside-0.4.1.tar.gz
| Download URL | benchside-0.4.1.tar.gz |
|---|---|
| Size | 216.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a65c7759ca75ef9d4d0ca7dfd1a01836283445dd9c1cc98b17cf84f6233c4042
|
|
BLAKE2b-256 checksum How to use checksums |
afbda441e7de3d94885dea97cd7f0a9dfbabd71dc2515cc48b8de0ef55e53de2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 6, 2026.
Transparency logRelease files / benchside-0.4.1-py3-none-any.whl
| Download URL | benchside-0.4.1-py3-none-any.whl |
|---|---|
| Size | 250.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
d339530506d435e3f535dba8646e7861d73c3c1d24319ed33293d5e18faab0db
|
|
BLAKE2b-256 checksum How to use checksums |
e629fac581d8c434314d4f6a1c2add24355b8b55b78f89e6053323ab13fd6ffd
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 6, 2026.
Transparency logSecurity for benchside 0.4.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for benchside-0.4.1.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 6, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
