Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.termyte.xyz/llms.txt

Use this file to discover all available pages before exploring further.

​
Termyte v0.1

“Stateless agents repeat mistakes. Termyte agents learn from them.”
Termyte is a safety runtime for AI coding agents. It provides a persistent layer of governance that redacts secrets, gates dangerous commands, and builds a “failure memory” so your agents stop repeating the same mistakes.

​
The Problem: Stochastic Looping

Coding agents often get stuck in loops—trying the same broken command or deleting files they shouldn’t. Even with large contexts, agents lack a persistent “muscle memory” for what works and what fails in your specific project.

​
The Solution: Explicit Governance

Termyte introduces a Safety Runtime that sits between your agent and your system. It works via three core tools that your agent uses to verify actions before execution:
  1. context_build: Initializes the task context and loads historical failure patterns.
  2. guard_action: Verifies risky non-shell intents (file deletions, git mutations) against project policies.
  3. execute: Runs shell commands through a governed pipeline with secret redaction and pattern matching.

​
Core Features

  • Failure Memory: Every agent failure is recorded. The next time an agent tries the same approach, Termyte warns or blocks it, forcing a strategy change.
  • Local Redaction: Secrets and PII are redacted locally by the MCP bridge before any data is sent to the cloud ledger.
  • Safety Policies: Define exact strings or regexes that your agent is never allowed to execute (e.g., rm -rf /, git push --force).
  • Audit Ledger: A complete, searchable history of everything your agent attempted and the outcome.

​
Getting Started

Install the Termyte MCP server and initialize your project governance: 
npx termyte init
Quickstart | Why Termyte? | Architecture