The objection we hear first
Nearly every conversation about AI with a law firm reaches the same point, usually within the first ten minutes. Someone says a version of this: our documents are in iManage, they are governed there, and we are not copying them somewhere else so a vendor can index them.
That is the correct instinct. A firm's DMS is not just storage. It carries matter structure, ethical walls, retention schedules, security classes and a decade of decisions about who can see what. Lifting documents out of it does not just create a second copy to secure. It leaves every one of those decisions behind.
So our integration starts from the opposite premise. iManage stays the system of record. AtlasAI works on top of it, and nothing moves.
What the partnership is
AtlasAI has joined the iManage Elevate Partner Program under an Alliance Partner Agreement, with access to the iManage Work Universal API. Elevate is the program for partners building products that integrate with iManage for the benefit of shared customers, and it is the route by which our connector becomes something a firm can turn on rather than something a vendor bolts on.
The integration runs on the iManage Work Universal API. Documents, folders and profile metadata come in. Work product goes back out as new documents or versions with their profiles set. iManage remains the authoritative repository for storage, governance, permissions, retention and access control, exactly as it is today.
Access follows the firm, not the vendor
The part that matters most is also the least visible. Every call AtlasAI makes to iManage is made as the signed in user, through OAuth 2.0, against the permissions that user already has.
If an associate cannot open a document in iManage, they cannot reach it through AtlasAI. If a matter is behind an ethical wall, that wall holds. If a security class changes on Tuesday, the change is live in AtlasAI on Tuesday, because AtlasAI never had its own separate answer to the question.
This is a design decision with real consequences. It means we cannot offer a firm-wide index that quietly ignores permissions, and we would not want to. Any product that can answer a question about a document you are not allowed to read has already failed the only test that matters in a law firm.
Where the graph comes in
Retrieval alone does not get a firm very far. Search a DMS for "indemnification cap" and you get documents containing that phrase, ranked by relevance. That is useful, and it is not the question anyone actually asked.
The question is usually relational. What positions have we taken with this counterparty. Which of our own precedents departed from the house position, and why. Who negotiated the last three deals in this structure.
AtlasAI builds a knowledge graph over the firm's content: parties, counterparties, matters, clauses, defined terms and the relationships between them. Entity resolution collapses the eleven spellings of one counterparty into one node. A counterparty backbone links matters to the entities on the other side of them. Clause level structure means a question about indemnification is answered from indemnification provisions rather than from whole documents that happen to mention the word.
Applied to iManage content, that turns the DMS from a place where documents are stored into something the firm can ask relational questions of. Same documents, same permissions, different question surface.
What that gives an attorney
The graph is infrastructure. What a firm actually uses is the set of tools built on it, all of which operate on iManage content in place.
Enterprise search combines keyword, vector and graph retrieval, narrows to the specific passage, and cites the source document so the answer can be checked rather than trusted.
Chat answers questions against the firm's own matters with citations back into iManage, not against a general model's memory of the internet.
Drafting works from the firm's own precedent. Playbooks compare a counterparty's draft against the firm's approved positions and produce a genuine tracked change redline in the original file, in Word or in PDF, with margin comments addressed to the other side.
Tabular review asks the same question of every document in a set and returns a grid with a citation in each cell, which is the shape diligence and lease abstraction actually take.
Workflows and agents plan a multi step task, run it across a matter, and show their working at each step.
The Word add-in puts search, precedent and redlining in the document the attorney already has open, which is where most legal work happens and where most legal AI is not.
What comes next
The graph gets better the more of a firm's history it sees. Every matter that lands in it makes the next answer sharper, because the precedent base and the counterparty history are the firm's own rather than a vendor's corpus.
That is the compounding part, and it is the reason to connect the DMS rather than upload a folder. A firm that has run on iManage for a decade already owns the most valuable legal dataset it will ever have. It has simply never been queryable as one thing.
If your firm runs iManage, this is a good conversation to have now. The shape of the first deployments will inform what we build next.
The shorter version
Your documents stay in iManage. Your permissions stay in iManage. Your retention and governance stay in iManage. AtlasAI reads what a given person is allowed to read, builds a graph of how it all relates, and puts search, chat, drafting, review and the Word add-in on top of it.
No migration. No second copy of the DMS to secure. No new permission model to keep in sync with the one you already maintain.