OpenSETI Agent Network

OpenSETI Agent Network is an open-source volunteer project for auditable SETI candidate and astronomy data review tasks. It supports two participation modes:

• BYOK API Worker: volunteers run a local worker with their own API key.
• Agent Client: volunteers open this repository in Codex, Claude Code, or a similar local coding agent and complete a task with human review.

The coordinator only distributes public tasks and receives auditable results. It does not collect user accounts, OAuth tokens, API keys, browser sessions, or subscription access.

Homepage: https://openseti.chaochaoweb3.com

For a Chinese architecture walkthrough, see docs/architecture.zh-CN.md.

For project direction and maintainer workflow, see ROADMAP.md and docs/maintainer-operating-loop.md. For release confidence, use docs/release-readiness.md. For new task intake, use docs/task-intake-rubric.md.

Project homepage source lives in site/ and can be hosted by GitHub Pages or any static web server. The architecture can later support other scientific review tasks, but the public v1 focus is SETI and astronomy.

What This Is For

The first demo uses public astronomy/SETI-style candidate review. The goal is not to claim alien discovery. The goal is to make public scientific review work easier to reproduce:

• summarize candidate signal metadata
• cross-check target context
• classify likely interference or instrument explanations
• produce structured JSON and Markdown evidence
• let multiple volunteers independently review the same task

Quick Start

git clone https://github.com/chaochaoweb3/openseti-agent-network.git
cd openseti-agent-network
python3 -m venv .venv
. .venv/bin/activate
pip install -e ".[test]"
pytest

Run the coordinator:

python -m osan.server --host 127.0.0.1 --port 8765

Run one dry-run worker task:

scripts/run-task --task tasks/openseti-demo-001.json --output results/demo-result.json --dry-run
scripts/check-task-intake
scripts/validate-task-manifest
scripts/validate-task tasks/openseti-demo-001.json
scripts/validate-result results/demo-result.json

Run the local release smoke path in one command:

scripts/release-smoke --repeat 3 --task-id openseti-demo-001

Submit the result to the local coordinator:

curl -X POST http://127.0.0.1:8765/v1/results \
  -H 'content-type: application/json' \
  --data-binary @results/demo-result.json

View the task catalog, local leaderboard, and one task review summary:

curl http://127.0.0.1:8765/v1/tasks
curl http://127.0.0.1:8765/v1/leaderboard
curl http://127.0.0.1:8765/v1/tasks/openseti-demo-001/summary

Run continuously from the coordinator:

scripts/run-task \
  --coordinator-url http://127.0.0.1:8765 \
  --submit \
  --repeat 0 \
  --interval 1 \
  --stop-after-seconds 3600 \
  --max-total-cost-usd 0 \
  --dry-run \
  --worker-id local-volunteer

Use --repeat 0 for an infinite loop. Start with --dry-run before spending API budget. For real API providers, set --max-total-cost-usd to a real budget and use a conservative --interval.

BYOK API Worker Mode

Dry-run mode is the default safe demo path and does not use any paid API.

To use an API provider, configure the key only on your own machine:

export OPENAI_API_KEY="..."
scripts/run-task --task tasks/openseti-demo-001.json \
  --output results/openai-result.json \
  --provider openai \
  --model gpt-4.1-mini

The worker reads API keys from local environment variables and never writes them to result payloads. Result validation rejects common secret fields and token patterns before submission.

Agent Client Mode

Open this folder in Codex or Claude Code and ask:

领取一个 OpenSETI 任务并生成结果报告。

The agent should follow AGENTS.md, inspect tasks/*.json, produce a result matching schemas/result.schema.json, run scripts/validate-result, and leave submission under human control.

Security Boundary

This project intentionally avoids centralized credential pooling.

• Never upload API keys, OAuth tokens, cookies, session IDs, or browser profile data.
• Never automate ChatGPT or Claude web subscriptions as a remote proxy.
• Never sell, transfer, or broker another volunteer's account access.
• Keep model credentials local to the volunteer machine or volunteer-owned deployment.
• Do not claim discovery of extraterrestrial life from demo outputs. LLMs help organize, explain, review, and prioritize evidence; humans make scientific judgments.

Current Status

This repository is a working prototype. It includes:

• a local coordinator API
• a deterministic dry-run worker
• optional OpenAI and Anthropic BYOK provider paths
• a Codex/Claude Code agent workflow
• result schema validation and secret scanning
• task intake validation, scoring, and a public task rubric
• deterministic task manifest checksums for public fixtures
• task catalog and per-task result aggregation endpoints
• a synthetic SETI candidate review demo
• a public TESS-SPOC metadata-only transit review fixture

It does not yet process raw radio telescope data or make autonomous scientific claims.

Project Direction

The next milestones are documented in ROADMAP.md. Maintainer agents should follow the operating loop in docs/maintainer-operating-loop.md: refresh live state, classify work, pick one primary outcome, verify it, and leave a clear next step. Before public announcements, run the checks in docs/release-readiness.md.

Acknowledgements

This project was partly inspired by ideas discussed by X user @disksing. Mentioning this inspiration does not imply endorsement, affiliation, or responsibility for this prototype.

License

Code is licensed under Apache-2.0. Example task/result data is intended for public scientific reuse under CC0 unless a task declares stricter source terms.