Agent: give your AI agent a program, not a mouse

OpenAdapt Agent gives an AI agent on your computer its compiled programs as tools, so it can list them and read the result. Your agent hands the values it chose to a program, which enters and checks them without calling a model. An operator first has to let the agent run programs. It ships in the openadapt launcher 1.17.0.

Updated 2026-11-19

What it does

  • List approved programs and the values each needs, without exposing recorded data.
  • Run an approved program, once an operator turns that on; the engine applies its own checks.
  • Read a run's result. A run that stopped is never reported as a success.
  • Package a program as an Agent Skill for AI assistants.
  • Turn an agent's successful run into a program, so later runs make no model calls (tested on synthetic records).
  • Let a named agent answer saves under a grant that a person signs.
  • Check a save made by any agent or script against the saved record (verify-run).

Status

Agent is released on PyPI as openadapt-agent 2.2.1, with an MCP server and Agent Skills. Start it with openadapt agent serve (openadapt launcher 1.17.0).

See it on your process

Bring a task your team types every day. We'll tell you whether it fits and what a test on your own system would cover.