Skip to main content

​
Give coding agents room to work without giving up control

Termyte is a local-first safety runtime for coding agents and risky developer commands. It checks known dangerous actions, applies your policy, records decisions, and remembers commands you explicitly marked unsafe. Start by checking command text without executing it: 
termyte check "cat .env"
termyte check "git push --force origin main"
termyte check "npm publish"
Termyte explains whether each command is allowed, needs review, or is blocked.

Check before execution

Evaluate risky command text without running it.

Set local policy

Apply built-in, user-level, and repository-level rules.

Set up an agent

Prepare Codex, Claude Code, or Aider with Termyte context.

​
Why use Termyte?

Coding agents move quickly and inherit your local permissions. A mistaken command can expose secrets, rewrite shared Git history, delete source, publish a package, or change infrastructure. Termyte adds a deterministic decision layer that does not depend on agent reasoning or an LLM: 
command -> understand action -> inspect targets -> check risk and policy
        -> allow, warn, ask, or block -> record the decision
You get:
  • consistent guardrails across supported agents and command checks;
  • built-in protection for recognized high-impact actions;
  • global and repository policy you can review in YAML;
  • local logs and user-controlled safe/unsafe memory;
  • experimental governed execution for commands and shell sessions;
  • no signup, cloud dependency, or LLM in the decision path.

​
Choose the right workflow

GoalUse
Safely see how Termyte classifies a commandtermyte check "<command>"
Test policy without writing a check logtermyte policy test "<command>"
Configure global or repository rulestermyte policy ... and YAML policy files
Launch a supported coding agent with Termyte contexttermyte run <agent>
Run one command through the experimental runtimetermyte run -- <command>
Start an experimental intercepted shelltermyte shell
Verify local runtime readinesstermyte doctor
termyte run &lt;agent&gt; currently launches agents in limited mode. It prepares local policy, logs, memory, and session context, but does not guarantee that every agent subprocess is intercepted. Use termyte shell only when you want to evaluate the experimental interception runtime.

​
Start here

Quickstart

Install Termyte and prove the core workflow in a few minutes.

How governance works

Understand decisions, policy layers, memory, and runtime boundaries.