AI in OCM
Everyone has a chatbot now. The question is what it can see.
OCM ships two, and keeps them apart on purpose. Casey answers how the software works and never sees a client file. Matter AI answers questions about the matter in front of you, by retrieving that matter’s own documents, and it shows you which ones it read, so an advocate can check the work before relying on it.
- Problem
- 62 · Eviction / ejectment
- Funding
- LSC basic field
- Advocate
- R. Ellsworth
- Status
- Open · merits
Ask the matter
What court is the hearing in, and what have we filed?
The hearing is set in Courtroom 4B of the Circuit Court for Jones County, before Judge Smith. An answer and a jury demand were filed on July 9.
Drawn from
- notice_of_hearing.pdf
- answer_filed_0709.pdf
Two assistants, deliberately separate
One knows the software. One knows the file. Neither pretends to be the other.
Casey, the helper
Grounded in the manual
Sits in the corner of every screen and answers “how do I close a case with an outcome code?” from OCM’s own documentation, linking to the page it came from. It deliberately never sees case content, and there is no reason a how-do-I question should carry a client’s file with it.
Matter AI
Grounded in the case
Lives on the case-detail screen and is built to read the file: metadata, activity timeline, parties, and the text of the documents attached to the matter, indexed per case so retrieval stays inside that case.
What it does
Eight places the assistant does actual work
Not one chat box. Each of these is a task legal aid staff already do, with the tedious half removed.
Ask the matter
A free-form question answered from the case's metadata, its activity timeline, its opposing parties, and retrieved excerpts of its own documents. Answers stream as they are written, and name the documents they drew on.
Summarize
Where the matter stands right now, in a few sentences, leading with status and the most recent meaningful activity, and flagging upcoming dates.
Draft a client email
A short instruction, such as status update, request medical records, or extension granted, becomes plain prose with no preamble, ready to edit and send.
Draft a motion
Free-form motion drafting from the facts already on the case, with a preflight check before it runs and a .docx filed to Documents only when an advocate accepts it.
Priority to-do list
One click turns your own open caseload into a prioritized action list, saved to your account, with any item convertible to a tickler. Dismissible if you would rather not see it.
Chronology
A dated narrative of what has happened on the matter, assembled from the record rather than typed by a paralegal the night before a hearing.
Semantic conflict check
Conflict searching that catches the near-matches an exact string search misses: a maiden name, a transposed name, a business the client is behind.
Transcription and translation
Recorded interviews transcribed onto the case, and translation for programs serving clients in more than one language.
See the work
AI features inside the case workflow
These are recorded in the configured demo system with synthetic matters. Pick a task to see where it starts, what evidence it uses, and what the advocate controls.
Ask the Matter
Real product footage · Captions included
Guardrails
The part that should decide whether you trust it
Confidentiality is not a feature request in legal services. It is the license. These constraints are in the product because the alternative is not defensible to a bar association.
Off until you turn it on
Every AI feature ships behind a toggle, default off. An organization that wants no AI at all runs OCM with none of it, and nothing in the product nags about it.
Consent before case data
Case content does not reach an AI provider until your organization has explicitly authorized it. This is a decision your board and your funders get to make, not a default we make for you.
Provider choice, including on-premises
Use Gemini on Vertex AI, another OpenAI-compatible service, or Ollama on hardware you control. That lets a program choose between a managed provider and inference that stays inside infrastructure it controls end to end.
Your own AI project
Each organization's AI runs in a cloud project provisioned for it alone, with a hard budget cap that disables only that project if it is ever exceeded. No pooled quota, no shared context.
Ethical walls hold
The document retrieval filter is enforced on the server, not read from the request. A user walled off from a case cannot retrieve that case's document text even by asking for it directly.
Proposed, never applied
Generated content is offered with a save button. Nothing an AI produces is written into the case record without a person choosing to put it there.
Honest limits
What we will not claim
A vendor that only tells you what its AI is good at is telling you half of what you need to make the decision.
AI output is a draft prepared by a tool, not legal advice and not a substitute for an advocate’s judgment. OCM is built so a person is always the one who commits it to the record.
We also monitor quality rather than assuming it: administrators get an AI health view showing what is being asked, what is failing, and what it is costing.
Ask it something about a real matter
The fastest way to judge a legal AI feature is to put a hard question from your own caseload to it and see whether the answer cites anything.