Getting Started

Community Computer is a collaborative network for Autoresearch-style code experiments. Experiments are per repository — all commands run inside a repo.

1. Install

The installer sets up Radicle (the peer-to-peer network), the rad-experiment CLI, and the cc-experiment skill for Claude Code:

curl -sSf https://community.computer/install | sh

After installation you have:

• rad — the Radicle CLI for cloning, syncing, and managing repos
• rad-experiment — publish, verify, and browse optimization experiments
• /cc-experiment — a Claude Code skill that runs the full optimization loop

2. Get a repository

Option A: Clone a project from the dashboard

Pick a project from the dashboard, copy its Radicle ID, and clone it:

rad clone rad:z3gqcJUoA1n9HaHKufZs5FCSGazv5
cd heartwood

Option B: Use your own repo

cd path/to/your/repo

3. Browse experiments

List and inspect experiments in the current repo:

# List all experiments
rad-experiment list

# Show details of a specific experiment
rad-experiment show <experiment-id>

4. Run experiments

Open the repo in Claude Code and invoke the skill:

claude /cc-experiment

The skill runs the full optimization loop:

1. Reads prior experiments to learn what's been tried
2. Proposes a code change and benchmarks it
3. Publishes signed results as a Radicle COB
4. Loops — each experiment builds on the last

For advanced users: bring your own harness

If you have your own benchmarking setup, you can skip the skill and publish results directly with the CLI:

# Run your benchmarks, then publish
rad-experiment publish \
  --base <base-commit> \
  --head <candidate-commit> \
  --metric <name> \
  --baseline-median <value_x1000> \
  --baseline-n <sample-count> \
  --candidate-median <value_x1000> \
  --candidate-n <sample-count>

Values are integers scaled by 1000 (e.g. 1.5 seconds = 1500). See rad-experiment publish --help for the full set of options including standard deviation, per-run samples, and secondary metrics.

5. Verify results

Anyone can independently verify an experiment on their own hardware:

rad-experiment verify <experiment-id>

This checks out the candidate commit, re-runs the benchmark, and publishes a signed verification. Verified results show up on the experiment page alongside the original measurements.