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.

​
Governance Tools

Termyte v0.1 exposes a single, high-integrity tool to the AI agent. All governance (safety checks, redaction, and logging) happens invisibly within this tool’s execution flow.

​
1. execute

Role: Mandatory Execution Broker This is the primary interface for all shell-related actions. Instead of giving your agent direct access to a “shell” or “terminal” tool, you provide execute.

​
Parameters

ParameterTypeRequiredDescription
commandstringYesThe binary or command to run (e.g. npm, git, python).
argsstring[]NoList of arguments to pass to the command.
cwdstringNoThe working directory for execution.

​
The Invisible Lifecycle

When an agent calls execute, Termyte performs the following steps automatically:
  1. Sanitization: Scans the command and arguments for secrets (API keys, tokens) and redacts them locally.
  2. Governance Prepare: Sends a request to the Termyte Cloud with the sanitized intent.
  3. Verdict Check:
    • If ALLOW: Proceeds to execution.
    • If BLOCK: Returns a failure message to the agent explaining why the action was stopped.
  4. Sandboxed Execution: Runs the command in a secure environment using execFile to prevent shell injection.
  5. Governance Commit: Streams the outcome (success/failure, exit code) to the Termyte Ledger for future learning.

​
Example Call

{
  "name": "execute",
  "arguments": {
    "command": "npm",
    "args": ["install", "uuid"]
  }
}

​
Technical Note: Manual Tools

While execute is the recommended way to use Termyte, the underlying API supports granular tools like context_build and causal_record. These are currently reserved for internal use by the MCP Bridge to ensure “Invisible Governance” but can be exposed for custom integrations in the future.
Quickstart | Architecture