We're Tisha and @susheem-k, and we maintain two open-source projects for people building AI agents. Both are opted into Hacktoberfest 2026 with good first issues tagged and waiting.
The projects
TokenOps — cost and control governance for AI agents. Budgets, guardrails, a spend ledger, and policy enforcement across single- and multi-agent runs. If you've ever watched an agent quietly burn through your token budget, this is the layer that stops it.
Chronicle — record-and-replay for agents. Capture a production failure once, then re-run your fix deterministically without live model calls, turning an incident into a committed regression test. (To appear at REALM @ EMNLP 2026.)
Where to start
We've tagged self-contained issues you can actually finish in a sitting:
TokenOps
- Node.js client wrapper and .NET client wrapper
- Additional LLM SDK adapters
- Implement the CANCEL actuator
Chronicle
- Per-provider request/response adapters
- Test coverage for StructuralAssertions
- A pytest helper: trace → one-line regression test
Full list: the hacktoberfest label on each repo.
How to contribute
- Comment on an issue before you start, so we can agree on the approach.
- Keep PRs focused, add tests when behavior changes, run the checks.
- Sign off your commits (DCO).
We review for quality, not volume, so low-effort or automated PRs get marked invalid. But a real, reviewed PR here is a genuine contribution to tooling that keeps AI agents safe and affordable, and we'll help you land it.
See you in the issues.
Top comments (0)