Plural
Plural is a Python SDK and CLI for finding out which AI is good at your work. You describe the work once, run any model against it, and keep the evidence behind every score.
The idea, in eight words
- App: one system the AI works with, such as a game written in Python, an MCP server, or an HTTP API. What its actions return is all the AI sees; its State stays hidden.
- Environment: the world, made of Apps, plus where it runs and its limits.
- Task: one assignment in an Environment, in plain language.
- Verifier: the grader. Its
checkscores the finished work, and it can give rewards along the way for training. - Agent: a model plus its instructions.
- Benchmark: a versioned set of Tasks, ranked by score.
- Job: one run of Agents on a Task or Benchmark.
- Trial: one Agent on one Task, once, with its full record.
Several Agents can also share one Task, to cooperate or compete: give the Environment
a party, and each Agent plays from its own seat with its own view and score.
Only what an action returns ever reaches the model. Scores, rewards, App State, credentials, and secrets never do.
Install
Plural requires Python 3.10 or newer.
pip install "plural[mcp]>=0.21.0"
Try a finished project
The support queue includes an Agent that follows keyword
rules instead of calling a model, so it runs offline with no account. From a copy of
examples/first-project:
plural benchmark validate support-triage
plural run --benchmark support-triage --agent scripted
plural job show <job-id>
Start your own
plural project init my-eval
cd my-eval
plural env init support-desk --python
plural verifier init resolved
plural task init refund --environment support-desk --verifier resolved
Each init writes a template. Fill in the places marked PLURAL-TODO, then:
plural task validate refund
plural auth login --api-key-stdin
plural run --task refund --model openai/gpt-5.6-luna
Every model call goes through the Plural gateway, billed at the provider's price, so a
real run needs a Plural API key from pluralintel.com. Every
run is a Job, recorded under .plural/jobs/ with each input pinned by version and
content hash, so plural job rerun repeats it exactly.
To run on Plural's servers instead, connect the folder to a private hosted project and
add --hosted:
plural project init my-eval --push
plural run --task refund --model openai/gpt-5.6-luna --hosted --follow
Pushing never makes anything public.
Python
The CLI and the SDK load the same resources:
from plural import Job
from plural.project import Project, Workspace
workspace = Workspace(Project.find())
benchmark = workspace.get("benchmark", "support-triage")
agent = workspace.get("agent", "scripted")
result = Job(benchmark, agents=[agent]).run()
Learn more
Start with Getting started, or read Core concepts first. Coming from 0.20? The changelog lists every breaking change and what to write instead.
Development
uv sync --group dev --group docs
uv run ruff check .
uv run mypy --strict src/plural tests/typing/consumer.py
uv run pytest
uv run python scripts/check_docs.py
uv run mkdocs build --strict
License
Apache-2.0
1 maintainer
Metadata
1 maintainer
Release files for plural 0.21.0
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 | |
|---|---|---|---|---|
| plural-0.21.0.tar.gz | 1.0 MB | With provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| plural-0.21.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 1.6 MB
Release files / plural-0.21.0.tar.gz
| Download URL | plural-0.21.0.tar.gz |
|---|---|
| Size | 1.0 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
25402f5048d3b3a914189e1db8107b47c764a6ca5e3068d00938b6238d77e858
|
|
BLAKE2b-256 checksum How to use checksums |
0be2c6c1fd5f994d894601d76e0fb9a8eeb4e241e699584971beb0175a80f68b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
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 Oct 13, 2026.
Transparency logRelease files / plural-0.21.0-py3-none-any.whl
| Download URL | plural-0.21.0-py3-none-any.whl |
|---|---|
| Size | 605.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
43a62cd8889b0b03cc9967e503a253f608cc61be232f2d7ca869e648ad1440a3
|
|
BLAKE2b-256 checksum How to use checksums |
17b960f63c6446eede5ca2d77e8239a33523c76ae5aca62f54c49c92ec806f33
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
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 Oct 13, 2026.
Transparency logSecurity for plural 0.21.0
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
All 2 files in this release have been uploaded with provenance metadata.
Attestation for plural-0.21.0.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 Oct 13, 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.
