Comments
>log in to comment
Maker here. I built punchtape after too many evenings of checking work the agent had already called done. The words you give are recorded word for word, the machine runs every check itself, and the finish is sealed only when everything stands. Four recorded demo sessions (Go, Rust, Python, Node) are in the repo - you can replay them without installing anything.
More AI tools →80 free dev tools →
- State machine ensures AI completes tasks end‑to‑end with verified claims
- Works with any IDE or terminal (Cursor, Copilot, Claude Code, Codex CLI)
- Fully offline, open‑source MIT license, no account needed
- One‑line install via curl script
punchtape is an open-source state machine that verifies AI-generated code until every check passes, ensuring true completion.
- for
- Developers using AI assistants to write or modify code who need reliable verification.
- pricing
- free
- license
- MIT
Key features
- External transactional state machine — Runs alongside your project, independently judging AI work and recording finishes.
- Real-time checks — Executes defined commands and compares exact output to verify each claim.
- Immutable task record — Stores words, checks, results and finishes in .punchtape/ for full auditability.
- Language-agnostic — Works with any programming language, framework or agent that has a shell.
- Offline and no account — Runs locally on Linux/macOS without network calls, accounts or telemetry.
- Incremental re-checks — Only re-runs checks affected by changed files, saving time.
Use cases
- Validate a new feature added by an AI assistant before merging
- Ensure an existing repository meets defined behavior after AI refactoring
- Automate bug detection and fixing loops for AI-generated patches
- Maintain deterministic builds across different AI models
punchtape vs alternatives
LangGraph | Mastra | LangChain | |||
|---|---|---|---|---|---|
| Best for | Transactional verification of AI-generated code | Stateful agent orchestration | TypeScript AI agent development | Autonomous code generation | LLM app construction |
| Pricing | Free | Free | Free | Subscription | Free |
| DevHunt upvotes | 1 | 0 | 0 | 0 | 0 |
| Launched | Nov 2026 | — | — | — | — |
- punchtape vs LangGraph: LangGraph provides a framework for building stateful agents but runs inside the agent, not as an external verifier.
- punchtape vs Mastra: Mastra is a TypeScript framework for AI agents, focusing on development rather than external verification of output.
- punchtape vs Devin: Devin is an autonomous AI software engineer that writes code but lacks a deterministic external check system.
- punchtape vs LangChain: LangChain is a library for building LLM applications, not a tool for enforcing task completion via checks.
punchtape FAQ
How does punchtape know when a task is truly done?+
The AI must output `done: true`; punchtape then re-runs all checks and records the finish only if every check passes.
Do I need to install anything besides the binary?+
No, just the single binary and a Unix shell; installation is a one-line script that places the command in ~/.local/bin.
Can punchtape work with existing projects?+
Yes, you can import an existing repo, write checks for its behavior, and run the machine to verify compliance.
Is any data sent to the cloud?+
No, punchtape operates entirely offline; all records are stored locally in the .punchtape folder.
What platforms are supported?+
Linux and macOS on amd64 or arm64; Windows is not supported.
Summarized by DevHunt from github.com · Nov 16, 2026. Details may change; check the official site.
LangGraph
Mastra
LangChain




