Skip to content

Repository files navigation

BenchFlow

The universal environment framework — a benchmark is just a frozen environment.

PyPI package Discord

What

BenchFlow is a universal environment framework: it runs AI agents against task environments and scores them through one hardened contract. A benchmark is just a frozen environment — point BenchFlow at any of them, drive it with any ACP agent, and run single-agent, multi-agent, or multi-round patterns over the same Scene-based lifecycle.

Qucik start: 1. Upload a trajectory

For local Claude Code or Codex jsonl formatted trajectory files that you are proud of (contain challenging tasks you dealed with via chatting to agents), you can upload that file to BenchFlow to join the competition for winning $2,000 cash reward. No BenchFlow account, credentials, or API key is required:

# Install or upgrade BenchFlow
uv tool install --python 3.12 --upgrade benchflow

# Upload one capture
bench traj upload /absolute/path/to/trial \
  --github-id YOUR_GITHUB_ID \
  --email YOU@example.com

The path may be a trial directory, a directory of JSONL files, or one JSONL file. Both contributor fields are required and are stored in manifest.json. See the concise upload skill or the trajectory upload guide.

Qucik start: 2. Run with a ChatGPT or Claude subscription

No OpenAI or Anthropic API key is required. Start Docker, install BenchFlow, then run one of these options. BenchFlow detects the saved host login and makes it available to the agent inside the sandbox.

uv tool install --python 3.12 --upgrade benchflow
docker info >/dev/null  # Docker must be running

ChatGPT subscription via Codex

Install the Codex CLI, then:

codex login
unset OPENAI_API_KEY CODEX_API_KEY  # ensure subscription auth is used

bench eval run \
  --source-repo benchflow-ai/skillsbench \
  --source-path tasks/citation-check \
  --agent codex \
  --model gpt-5.5 \
  --sandbox docker

Claude subscription via Claude Code

Install Claude Code, then:

claude auth login
unset ANTHROPIC_API_KEY ANTHROPIC_AUTH_TOKEN  # ensure subscription auth is used

bench eval run \
  --source-repo benchflow-ai/skillsbench \
  --source-path tasks/citation-check \
  --agent claude \
  --model claude-sonnet-4-6 \
  --sandbox docker

The agent may pass or fail the benchmark task; either result means the evaluation completed. Each run writes rewards, token usage, and the full trajectory under jobs/. See Getting started for other agents, models, and sandboxes.

Install

Install or upgrade to the latest stable release from PyPI with uv:

uv tool install --python 3.12 --upgrade benchflow
  • Confirm with bench --version.
  • BenchFlow CLI releases require Python 3.12 or newer. Keep --python 3.12 in the install command so uv does not resolve an older Python-compatible package that lacks the CLI entrypoints.
  • If you see Executables already exist: bench, benchflow, re-run with uv tool install --python 3.12 --upgrade --force benchflow to replace stale entrypoints from an older install.
  • For Daytona, Modal, or AgentCore extras, install the relevant optional package, for example uv tool install --python 3.12 --upgrade 'benchflow[sandbox-daytona]'.

Internal users wanting the newest preview from main install the internal preview channel (uv tool install --python 3.12 --prerelease allow --upgrade benchflow).

Requirements & auth. Install uv; the --python 3.12 flag lets it provision a compatible interpreter for the tool install. Set DAYTONA_API_KEY for Daytona or configure Modal auth for Modal; export an agent API key (GEMINI_API_KEY, ANTHROPIC_API_KEY, …) or use subscription auth (claude auth login / codex login). Provider-prefixed models may need provider-specific credentials; Azure Foundry uses AZURE_API_KEY + AZURE_API_ENDPOINT.

Documentation

Start with Getting started, then Concepts for the mental model. Prefer to have an AI coding agent run the whole quickstart for you? Paste the agent quickstart prompt into Claude Code, Codex CLI, or Gemini CLI. Then by goal:

If you want to… Read
Run an eval on an existing task Getting started
Understand how BenchFlow runs any benchmark (the three-layer model) Run any benchmark
Have an AI agent install + run the quickstart end to end Agent quickstart prompt
Run an agent from the public agents repo (goose, qwen-code, prime-agent, …) Running external agents
Understand Rollout / Scene / Role / Verifier Concepts
Author a new task Task authoring
Author a task in the native task.md format Native task.md authoring
Run a hosted PrimeIntellect / Verifiers environment CLI reference
Multi-agent: coder + reviewer, simulated user, BYOS, stateful envs Use cases
Multi-round single-agent (progressive disclosure, oracle access) Progressive disclosure
Skill evaluation (when the artifact is a skill, not a workspace) Skill eval
Contribute a trajectory capture Trajectory upload
Understand the security model Sandbox hardening
Use public vs internal preview SDK releases Release channels
CLI flags + commands CLI reference
Python API surface Python API reference

Notebooks and runnable example scripts live under docs/examples/ so examples stay versioned with the docs that explain them.

bench agent vs bench eval adopt. bench agent list / bench agent show inspect registered AI agents (the solver programs like Claude Code or Gemini CLI). Onboarding a third-party benchmark into benchmarks/<name>/ is a separate workflow — bench eval adopt <source> scaffolds and drives the conversion, and bench eval adopt <name> --verify parity-gates it. (The legacy bench agent create|run|verify commands still work as deprecated aliases.) See the CLI reference for details.

Benchmark task sources

Benchmark datasets live in external Git repos and are referenced with two fields:

# benchmarks/harvey-lab/harvey-lab-gemini-flash-lite.yaml
source:
  repo: benchflow-ai/benchmarks    # GitHub org/repo
  path: datasets/harvey-lab/tasks  # optional subpath within repo
  ref: main                         # optional branch/tag
agent: gemini
model: gemini/gemini-3.1-flash-lite-preview

Run any benchmark via the CLI:

# From a YAML config (shipped with the repo)
bench eval run --config benchmarks/harvey-lab/harvey-lab-gemini-flash-lite.yaml

# Inline — mirrors the YAML source fields
bench eval run \
    --source-repo benchflow-ai/skillsbench --source-path tasks \
    --agent gemini --model gemini-3.1-flash-lite-preview --sandbox daytona --concurrency 64

Repos are cloned and cached locally under .cache/datasets/ on first use.

Hosted environments are another source type. Instead of a repo, pass --source-env with the environment's pinned source version to run an external PrimeIntellect / Verifiers environment on its own native harness — BenchFlow preserves the hosted identity (env_uid, hub_url) and still writes the shared rollout output contract. See the CLI reference for the full hosted-environment command shape.

Downstream projects should depend on the public PyPI release by default. For internal validation before the next public release, install or lock the internal preview channel with prereleases enabled; see Release channels.

Authoring tasks

A task is one task.md (YAML frontmatter for config + a markdown prompt body) plus environment/ and verifier/ sidecars. The bench tasks commands cover the authoring lifecycle:

bench tasks init my-task                 # scaffold a task.md package under tasks/
bench tasks check tasks/my-task          # validate (default --level structural)
bench tasks migrate legacy-task/ --remove-legacy  # convert old split packages to task.md
bench tasks export tasks/my-task out/             # write a compatibility export + loss report

See Native task.md authoring and the task standard.

Featured

Audience

Contributing

PRs welcome. Open against main. CI runs ruff + tests on every PR; please run ruff check . and pytest tests/ locally first.

Release channels are documented in Release channels. In short: merges to main publish an internal preview after CI passes, while a matching release tag publishes the public release.

License

Apache-2.0.

About

Research infra for creating RL environments, post-training, and evals.

Topics

Resources

Stars

322 stars

Watchers

2 watching

Forks

Releases

Packages

Used by

Contributors

Languages