Renting the model is one thing. Renting the whole platform is another.
Most legal AI is a black box on someone else's stack. The model is theirs, the infrastructure is theirs, and so is the platform your firm now runs on. You can use it. You cannot see inside it, you cannot change it, and you cannot take it with you. When the roadmap does not match your firm, you file a feature request and wait.
That is a strange thing to accept for something that is quickly becoming core infrastructure. Firms do not rent their document management system's source code from a vendor who alone decides what it will ever do. Legal AI should not be different.
AtlasAI runs on code your firm owns
AtlasAI is deployed as a single tenant inside the firm's own Azure environment. The firm's data stays in the firm's boundary, under the firm's keys, with the firm's access controls and audit. That much is table stakes for us.
The part that changes the game is ownership of the platform itself. The codebase your AtlasAI instance runs on is yours. Not a hosted seat on a shared product, but a platform your firm holds, in your own repository, that your own people can read and extend.
What ownership unlocks: your engineers, your tools, your repo
Because it is your repository, you are no longer limited to what a vendor ships. Your engineers can work directly against the AtlasAI codebase, and they can bring modern agentic development tools with them. Point Claude Code straight at your own AtlasAI repo and it can read the platform, extend it, and ship changes the way any serious software team now builds.
That means new agent workspaces for a practice group, a custom integration into a system only your firm uses, a workflow that matches how your associates actually work. Built on your stack, at your pace, without waiting in a vendor's queue behind every other client. The platform becomes something your firm develops, not just something your firm subscribes to.
A concrete picture
Say your litigation group wants a workspace that assembles a pleadings matrix a particular way, and your firm has a data source no vendor has ever heard of. On a hosted product, that is a support ticket and a maybe. On a platform you own, an engineer opens the repo, works with an agentic tool like Claude Code to build the workspace and wire in the source, tests it inside your tenant, and ships it. Days, not roadmap quarters. The capability lives in your firm from then on.
The platform was already built to be built on
This is not a bolt-on. AtlasAI exposes an MCP and REST surface so any AI tool can reach the firm's curated knowledge in a governed way. The curated ontology and the agent workspaces were designed as a foundation for others to build on. Opening the codebase to the firm's own engineers is the natural next step: the same platform, now extensible by the people who use it.
Extending it does not break the trust boundary
Ownership does not mean giving up governance. Everything still runs inside your tenant, on your keys, with the same access controls and audit trail. Changes your team makes are changes inside your own boundary. You get the freedom to shape the platform and the control that made you trust it in the first place. Those two things usually trade off against each other. Here they do not.
Rent the model. Own the platform.
Models will keep changing, and you should keep swapping in the best one. That part is fine to rent. The platform your firm runs on, the knowledge it holds, and the code that turns them into work, that is the part worth owning. Ownership is the difference between adapting the tool to your firm and adapting your firm to the tool. With AtlasAI, and now with your own engineers and tools like Claude Code working directly against your own repo, that difference is finally yours to keep.