Built for the long haul.
Agents are brilliant for an afternoon. Atlas ADE is for the months after that, when they need to remember what they did, check their own work and keep a project from sliding into entropy.
Where it came from
Atlas ADE was built originally as a platform to support agents in long-horizon development workflows, run locally. Long horizons are where agents struggle, so it grew the parts that kind of work needs:
- agent loops, evaluations and reviews
- tooling the agents call instead of improvising
- knowledge derived from a graph of the code, delivered just in time
- usage and burn data, so you can see where the work goes
- self-assessment, and a work list the agents author, order and execute
Why it stands on its own
Somewhere along the way it became clear the harness could stand on its own and help other people. It took about 500 hours of headaches to learn how to brace the weak spots that exist in agentic development: agents that forget, agents that re-derive what they already knew, and systems that slowly fall apart. Those 500 hours don't need to be yours.
What it costs
Less than a month of a Pro subscription to Claude or ChatGPT. That leaves your usage for your ideas and your projects, not for working out why your agents keep forgetting things and your systems keep drifting. See pricing.
Who's behind it
Atlas ADE is developed by Connor, a former NHS Digital Lead and now an independent developer.
Say hello at [email protected], or see Support if something isn't working.