Bexar County Criminal Defense
AI on the MP4
Most legal AI reads documents. The line it cannot cross is the one where the evidence stops being words and starts being a recording.
The report is the State’s account, not the case
You get appointed on a Bexar County matter, and the discovery is not a brief. It is a thumb drive: nine hours of body-cam video, a 911 call, a recorded interrogation, forty scene photographs shot at two in the morning. The police report sits on top of that pile, and a document-only legal AI reads the report beautifully. It ingests the affidavit, the typed witness statement, the offense narrative, and it summarizes them cleanly.
But the report is words written about a recording, by the people who held the camera. A tool that can only read the report can only ever read the State’s version of what the footage shows. It tells you the story. It never tells you the case.
What it means to read the footage itself
Here is the line, and it is the whole product. The Evidence Orchestrator does not read the report about the footage. It reads the footage. It opens the raw body-cam MP4 and watches it. It listens to the 911 audio and the station-house interrogation. It looks at the scene photographs as the primary evidence they are. The MP4, the WAV, the JPEG: the actual things the State recorded, not the prose written over them.
That is the difference between a tool that can summarize a discovery letter and a tool that can tell you the report calls the man combative while the body-cam shows him sitting still on a curb for six minutes. The multimodal read is what crosses the line. Google’s Gemini does the multimodal review, with OpenAI’s ChatGPT and Anthropic’s Claude part of the reading, and what comes back is grounded in what the recording holds, not in what the report claimed about it.
What the AI hands back
Point it at a matter on the Case Board and the read comes back indexed and built to be cited: a transcript of every spoken word with the speaker and the time it was said, a frame log of what the camera saw, and a plain-language description of each photograph. You jump from a finding to the second of footage it came from, the way you cite a page in a brief. Nothing is invented and nothing is summarized away. Where the camera shows an interaction, the record says interaction. Where it shows nothing, the record says so.
The reading happens under a cost cap, and every AI action is logged to an audit table, because a defense record has to be a thing you can defend the handling of. The machine reads the tape and writes what is on it. It does not decide what the silence at 0:41:12 means. You do.
Why the line is the access-to-justice line
A well-resourced office assigns a paralegal and a budget to watch every minute of tape. The appointed defender with a fourteen-matter docket day gets through what the hours allow and hopes the rest does not matter. That is not a difference in talent. In a body-camera world, it has become a difference in tools, and a difference in tools has become a difference in the defense a poor person actually receives.
Gideon promised a real lawyer, and a real lawyer in 2026 is one who has read the recording and not just the report about it. The line a document-only tool cannot cross is exactly the line the Sixth Amendment now runs along. The Evidence Orchestrator was built to put the lawyer on the right side of it.
The whole machine
Ten logins down to two
One subscription stands in for the stack a defense firm usually stitches together, so the practice runs on two things it already trusts: Google Workspace, where the files and matters live, and The IMC Machine, where the matter runs. LawPay and QuickBooks, the money rail, stay where they are, fed by the Machine. Casework, evidence, motions, vouchers, marketing, and events, in one place.
San Antonio and Bexar County. The Machine is built for here first. For another city, county, or region, the build-out is custom and carries its own setup fee.
