Free
$0
For one person trying Atlas on their own work.
- Open source under AGPL-3.0
- One agent per task
- 7 ready-made agents
- Runs on your machine
Introducing
Hand it the work. Get it back delivered.
100
tasks delivered. Not 100 conversations.
No new AI subscription.
Atlas runs on the plan you already pay for, within its limits.
We use Atlas to deliver our own work. Write what you need, and it plans, builds, checks and delivers, using the AI subscription you already pay for.

This is the real Atlas delivery workflow. Every step has a specialist, and work only moves on once it passes. Select a step to see who does it.
Same request, two routes. One waits at every hand-off. The other splits the work, builds it in parallel, checks everything and delivers.
RequestAdd GST invoices to the merchant app
Ready
2 hand-offs, 1 queue, 1 rework loop
Ready
No hand-offs, queues or rework
Illustrative timeline, not to scale.
A lean senior team steering specialised agents.
Instead of a large team, with every role staffed for the whole project.
Each step starts the moment the previous one is ready, and independent work runs in parallel.
Instead of work waiting in a queue between every role.
Every change is checked automatically and reviewed before it is delivered. Anything that fails is fixed and checked again, and anything unclear goes to a person.
Instead of testing near the end, mostly by hand.
AI effort matched to each task, with nothing spent where it is not needed.
Instead of the same expensive AI for every task, or no AI at all.
Cost, speed and quality of every delivery, tracked live.
Instead of weekly status reports.
Every decision and change is recorded with the work, so nothing lives only in someone's head.
Instead of know-how that leaves when people do.
Atlas comes with 24 specialists and 6 workflow templates. Pick a kind of work to see who steps in. In delivery workflows, each maker is paired with a reviewer, so nothing passes on its own say-so.
Each item runs on the AI you already pay for, and comes back finished for your approval.
92 of 100 ready for your approval
Illustration. Real speed depends on your code, your checks and your AI plan.
Real Atlas screens.




Atlas runs on your own server, with the controls a company expects.
Most AI work tools bundle the AI into the price. Atlas doesn't, so you pay for the delivery system and keep the subscription you have.
| A person directs the work | Finished work comes back | |
|---|---|---|
| Your own AI subscription | Parallel agent toolsFree to about $60 per user a month. Several agents at once, and you still direct each one. | |
| AI bundled into the price | AI coding assistantsAbout $20 to $40 per user a month. Help for a developer, one conversation at a time. | Autonomous coding agentsAbout $40 per seat, plus a team fee and usage. Finished changes, on the vendor's cloud. |
| Type of tool | Typical cost | Where the AI comes from | What you get |
|---|---|---|---|
| AI coding assistants | About $20 to $40 per user a month | Bundled, and heavy use can cost extra | Help for a developer, one conversation at a time |
| Autonomous coding agents | About $40 per seat, plus a team fee and usage | Bundled and metered | Finished changes, on the vendor's cloud |
| Parallel agent tools | Free to about $60 per user a month | Your own subscription | Several agents at once, and you still direct each one |
| Enterprise agent builders | About $500 to $6,000 or more a month, or usage credits | Bundled | Business process automation, not software delivery |
| Atlas | Free to start, then $19 per user a month | Your own subscription | Planned, built, checked and delivered, on your own systems |
Typical public list prices by type of tool, October 2026. Prices change, and each vendor's own page has the current figure.
$0
For one person trying Atlas on their own work.
$19per user a month
For teams. $15 per user a month when billed annually.
Let's talk
For organisations with their own security, hosting or integration needs.
Prices are in US dollars and exclude taxes. Your AI subscription is separate: each person uses their own Claude or GitHub Copilot plan, and Atlas work counts against that plan's limits.
No. Atlas runs Claude Code or GitHub Copilot under each person's own sign-in, or models through Ollama, and you choose which one each agent uses. You keep paying your AI provider directly, and Atlas work counts against that plan's limits.
Atlas breaks each request into small pieces of work, runs specialised agents on them in parallel and matches the AI effort to each task, so nothing is overspent. Every result is checked and reviewed before an engineer signs off the delivery, and the cost of each delivery is tracked so you see what the spend achieved.
No. Atlas delivers into your repositories, where your team reviews every change. When we build for you, Atlas runs on our systems and the result is deployed to your own AWS or Azure account. If you license Atlas, it runs on your own systems.
Your team does. Atlas raises each finished change as a pull request and a person approves it. Atlas doesn't merge changes by itself.
Not in one repository. Atlas works on one task at a time per repository and queues the rest, so changes don't collide. Tasks in different repositories can run side by side, up to the limit you set for each workflow.
Atlas runs on one server in your organisation, using Docker and Postgres. Each person also runs a small Atlas runner on their own computer, where their agents work. A task needs its owner's runner to be online. You also add your own lint, build and test commands to the workflow, so Atlas can run them.
No. Atlas runs on your systems, and your code goes only to the AI provider you choose. There's no third-party analytics in the app. Online licences send us a daily usage count, such as active users and version, and offline licences are available.
GitHub.com, Jira Cloud, and alerts to Microsoft Teams and Telegram. If you use something else, tell us and we'll scope it with you.
Yes. It's open source under the AGPL-3.0 licence, and you only need your own AI subscription, or local models through Ollama, to run it.
Atlas isn't a shared hosted service. We can set it up in your own cloud account as an Enterprise project.