Infrastructure for trustworthy agent work.
Coding agents are fast but opaque: they touch your repo, and you're left without proof of what ran, why it ran, or how to undo it safely. Tessariq builds the layer underneath — agents run in isolated workspaces, leave durable evidence, and hand back results you review like any other change.
It's also an agentic-engineering lab: an open workbench for figuring out what trustworthy autonomy actually looks like in practice.
⚠️ Pre-release. The first version of the core tool isn't out yet — it's landing soon. Watch tessariq for the release.
New to the project? Begin with tessariq — the core runtime. Reach for taskrail when you need structured, verifiable task execution on top.
Once released, install will be a one-liner:
brew install tessariq/tap/tessariq| Project | What it does | Stack | |
|---|---|---|---|
| 🧰 | tessariq | Git-native runtime for running coding agents in isolated workspaces, capturing durable evidence, and promoting results back into normal Git review. | Go |
| 🚉 | taskrail | Execution harness for AI agents: goals in, structured tasks out, verified results back. | Go |
| 🍺 | homebrew-tap | Homebrew tap for installing Tessariq's tools. | — |
- Sandboxed by default. Agent runs stay isolated until you choose to promote them.
- Evidence you can audit. Every run leaves durable, reviewable artifacts — not just a final diff.
- Git-native. Results flow back through normal review, not a separate console.
- Reproducible. Deterministic task execution and verification gates, so the same goals produce results you can trust and re-run.
Issues and PRs welcome on any repo. Questions or ideas → open a discussion or reach out at felix@fmueller.io.