A real-time research tool built around the bar's competence, confidentiality, and disclosure duties — not against them. Upload your motions, briefs, and orders; the platform builds a structured library you review and approve. Citation-locked so it cannot fabricate authority. Ships with disclosure templates calibrated to ABA Formal Opinion 512 and the major state bars. Built for solos, small firms, and pro se litigants — and the judges and ethics committees who will scrutinize them.
Reviewing this for your firm? Read the partners' objections page — Mata v. Avianca, work-product privilege, Rules 1.1 / 3.3 / 5.3 / 5.5, stale law, sanctions exposure — each answered in the vocabulary of the bar.
Private beta · Free permanently for pro se · Hearing-day pass from $39/mo · macOS in v2
How to think about it
Local Motion doesn't represent you. It doesn't argue for you. It doesn't generate legal advice. It is not a substitute for counsel — and where you are unrepresented, it is not a substitute for retaining one.
What it does is surface your own curated library — the cases you found, the holdings you wrote, the orders you filed, the facts you organized — at the moment opposing counsel pauses.
Every citation it shows you is one you (or your team) put in the library. The tool only finds it faster than you could on paper. The same as a binder, a memo, or a junior associate handing you a tab.
The work is still yours. The judgment is still yours. The faster part is ours.
What it does
Local Motion doesn't just look up a case. The moment opposing counsel pauses, it reads everything in your library — the cases, rules, prior orders, and facts you've prepared or imported — cross-references the citation they just made against your record, and frames a response built for this matter, not a textbook answer.
Every case, rule, prior order, and fact you've prepared or imported — searched the instant it's needed. The tool isn't reaching for a single case; it's reaching for everything in your library.
The case opposing counsel cited, the case you'd want to cite back, the rule that controls, and the prior order — if any — where this exact issue was already decided. All in the same answer.
Not a generic response. A line shaped by your specific record — pointing at the paragraph of the prior order, the exhibit, the date in the procedural history. Read aloud as written, or use it as your starting point.
The moment
In a remote hearing, a deposition, or an oral argument, you have seconds. Opposing counsel cites a case; the judge waits; you scramble through three binders looking for the page where you wrote down the response. Then you say something close-enough, and the moment is over.
Local Motion is what you'd have if you could afford a junior associate sitting next to you whose only job is to whisper the right citation at the right second.
What it does
A made-up citation, a stale fact, or a privacy slip would be catastrophic in court. Every feature here is shaped by that.
Hallucinations are structurally impossible — not blocked by a prompt instruction, but ruled out by how the system is wired. The AI's citation tools are typed to your current library; an ID it didn't get from you cannot pass the schema. If a case isn't in your library, it cannot appear on your screen.
The post-Avianca architecture, in detail.
From the moment opposing counsel pauses for a breath, a response is on screen in a few seconds — ready before the judge looks back at you.
Refine the wording of a holding during a recess. Add a new case you just remembered. Tweak a fact when you spot a typo. Saves are live the moment you click — no reload, no break in the loop.
Your microphone and hearing audio are captured locally in memory. By default, audio chunks are sent over HTTPS to the speech-to-text provider you configure (Groq, OpenAI Whisper, or any compatible host) for transcription — that's how transcription works. Once the audio is turned into text, the audio is discarded; only the transcript moves on to the LLM. For audio to stay 100% on-machine, set the STT provider to local (whisper.cpp; path wired today, full pipeline integration v0.2). Recording is off by default.
Use OpenAI, Anthropic, or any OpenAI-compatible provider — your key, your account, your choice. Point at a self-hosted endpoint on your own network for full privacy control. Switch in Settings without restarting. The citation-lock applies regardless of provider.
Preview Managed AI — a hosted, legal-tuned model that removes the BYO-key step entirely — is in private preview with design partners. General availability targeted for v0.2.
Drop in your motion, your brief, opposing counsel's filings, prior orders, deposition transcripts. The platform extracts every case, rule, factual assertion, and demand into a list you review one at a time. Accept what's right; edit what isn't; reject what doesn't belong. The library is built by you, just faster — and every entry has a name on it.
Every case citation you enter is checked against the CourtListener corpus — the Free Law Project's free, open citation database. Typos and unrecognized citations are flagged inline; a one-click link opens the case's treatment history. The check catches what the lock can't: typos, wrong reporter, or a name that doesn't match the real opinion.
Any case in your library can be flagged as overruled, limited, distinguished, or abrogated. When the agent surfaces a flagged case in a hearing, the panel renders it in red with a high-visibility badge — so you don't read superseded authority into the record. Today the flag is manual; automated detection from CourtListener treatments is on the roadmap.
Every advice card the agent surfaced is captured automatically. Open the Review window after the hearing, attach a thumbs-up / thumbs-down + note to each, then export the whole record as Markdown or JSON. Forces reflection (Rule 1.1 lives here), tunes your library over time, and creates an audit trail you control.
How it works
Drop in motions, briefs, prior orders, exhibits, deposition transcripts. The platform extracts authorities, holdings, rules, factual assertions, and demands — and presents each one for your review and approval before it can be cited.
Local Motion sits on top of your video window. It picks up everyone's voice automatically. No plugins, no virtual cables, no special setup.
What everyone says becomes text on your screen as it happens — your words and theirs. The tool watches for the pause that means it's your turn.
Not just one case. A line you can read aloud, framed by the rule that controls, the prior order in the matter, and the paragraph in your record that supports it. The judge is still looking at opposing counsel.
Wondering "isn't this hallucinating?" — that's the Avianca question.
A look inside
Every entry is a typed form you fill out. Add a case, edit a holding, write down what to say when this case applies. Save, and it's live in the tool — the next citation can draw on what you just wrote.
Drop in opposing counsel's PDF. The tool reads it, pulls out the cases and rules they're relying on, the facts they're asserting, and the relief they're asking for. You review each one. Accept what you want; the library updates immediately.
Privacy & correctness
A made-up citation in front of a judge is a sanctionable error. A leaked communication is a bar complaint. We built Local Motion so neither is possible.
Read the full privacy policyThe tool is structurally restricted to your library. The AI it talks to is not permitted to suggest a case that isn't already in your files — not as a matter of prompt instruction, but as a matter of how the system is wired.
Even if a future AI model tries to break this, the tool checks every answer against your current library before it lights up on your screen. Anything unfamiliar is rejected. You never see it.
Captured locally in memory. By default, audio chunks are POSTed to the speech-to-text provider you configure (Groq Whisper, OpenAI Whisper, or any compatible endpoint) for transcription — they have to be, because that's how transcription works. Once transcribed, audio is discarded; only the text reaches the LLM. To keep audio fully on-machine, configure local Whisper (whisper.cpp) or a self-hosted STT endpoint on your own network. The privacy policy describes the boundary precisely, and the vendor-exception analysis covers the privilege question.
Your AI provider account, your API key, your control. The tool talks only to the providers you've named in Settings — nothing else, nowhere else. No telemetry, no analytics, no surprises.
Disclosure templates
Every standing order requiring AI disclosure, every bar opinion on AI assistance, every judge who asks "what tool was that" — answered with prewritten language calibrated to ABA Formal Opinion 512 and the published guidance of the NJ, NY, CA, TX, and FL state bars. Drop the snippet into your filing or read it aloud. Your candor stays the same. You just stop scrambling for the right words. The Rule 3.3 framing is in the objections page.
DISCLOSURE OF AI-ASSISTED RESEARCH TOOL
Counsel for [PARTY] hereby discloses, pursuant to [STANDING ORDER / RULE / BAR OPINION], that during this proceeding counsel may consult a real-time research and preparation tool to retrieve authorities counsel curated in advance. The tool does not generate legal advice; every citation it surfaces was placed in counsel's library before this proceeding. All legal judgment exercised remains counsel's own.
"Your Honor, before we begin, counsel respectfully notifies the Court that I will be referring to a digital prep tool during this hearing. It is a structured library of authorities I prepared in advance of today; the tool retrieves from that library and does not generate legal arguments. Happy to elaborate if helpful."
"Your Honor, the assistance counsel referenced is a research tool that retrieves from a library of authorities counsel curated before this proceeding. It does not write briefs, draft pleadings, or generate citations; every authority it shows counsel exists in counsel's pre-prepared library. Counsel reviewed each authority before this hearing and is prepared to articulate counsel's own legal position regarding any of them."
CERTIFICATION OF AI-ASSISTED RESEARCH USE
The undersigned certifies that: (1) the undersigned consulted a research tool during the [hearing / deposition] on [DATE]; (2) the tool surfaced authorities from a library the undersigned curated in advance; (3) the tool did not generate legal advice or pleadings; and (4) the undersigned independently verified each authority cited.
Every authority comes from the library you reviewed and approved. Same posture as a binder or a clerk handing you a tab — well-settled practice.
The tool does not represent or advise you. We say so plainly in the Terms and in every template we ship.
Templates tracked against published guidance from the ABA and the major state bars, updated as new opinions and standing orders issue.
The citation-lock means cases can't be fabricated. It does not certify that a holding still controls. Independent verification is your duty; we say so in every template and in our terms.
Sample language above is illustrative of what every paid plan ships and is updated when bar opinions, standing orders, or court rules change. It is not legal advice and should be reviewed against your jurisdiction's current rules and your firm's standards before use.
Pricing
Free permanently for pro se. Day-pass for occasional users. Flat for solos. Firm-wide — no per-seat math — for small firms. Every paid tier brings its own AI keys (any OpenAI-/Anthropic-compatible provider). Managed AI is in private preview for design partners.
For people representing themselves in matters they didn't ask for.
For paralegals, students, moot court, occasional appearances, small-claims litigants.
For solos and small-shop attorneys who still do their own motion practice.
For 2–10 attorney firms. Firm-wide pricing — no per-seat procurement math.
Larger firms or regulated industries with deployment requirements outside the standard tiers.
custom provider)Reviewing for partner approval? Start with the objections page.
Contact usA hosted model fine-tuned on U.S. case law and procedure, running on Local Motion's infrastructure. Flat-rate; no OpenAI/Anthropic account to configure; no usage-based surprises on long evidentiary days. The citation-lock applies the same way — every cite still comes from your library. Audio handling is unchanged: audio goes to the STT provider you configure (Managed AI's STT, your BYO key, or local Whisper); it never reaches the LLM.
Status: in private preview with design partners. The current desktop app supports the underlying architecture (point at any compatible inference endpoint via the custom provider in config.json). General availability of the hosted endpoint, with billing, is targeted for v0.2. Ask about early access →
v0.2 · Pro roadmap
v0.1 is shaped for one lawyer, one matter, one hearing. v0.2 Pro adds the infrastructure that lets a firm scale a multi-decade library across many matters, run inference without configuring an AI provider, and plug in proprietary inference deployments. The three deliverables below are in scope, not yet built. As each lands we promote it from Roadmap to Shipped in the Implementation Status table — nothing here is claimed before it works.
Today the agent reads the entire library on every trigger. That works up to ~50–100 entries; past that, token cost climbs and recall degrades. v0.2 embeds your library locally (no audio or library content leaves the machine for embedding) and pulls only the top-K relevant entries per trigger — sub-millisecond, regardless of library size. The citation-lock applies to the retrieved subset, so the enum the model sees is tighter than today.
Skip the BYO-key step entirely. Local Motion operates the inference; you authenticate with the same activation key you already use; billing is flat-rate. In v0.1 this is in private preview with design partners — the custom provider lets you point at any compatible endpoint today, but the Local-Motion-operated hosted endpoint and the billing piece are v0.2 deliverables.
The custom provider today accepts Anthropic-format or OpenAI-format endpoints. v0.2 generalizes it: point at a proprietary inference URL — a fine-tuned model on your GPUs, a Bedrock deployment, a vLLM server — and configure the request/response mapping. The citation-lock applies regardless of upstream because the constraint lives in the tool-use schema, not the model.
Want early access, want to weigh in on scope, or want to use Local Motion at firm-scale before v0.2 lands? Get in touch. Detailed architecture and rollout plans are available under NDA on request.
Questions
Reviewing this for your firm? The questions a litigator asks are different — see the objections page.
Early access
We're admitting a small private-beta cohort in waves. Tell us who you are and when your access is approved we'll send your activation key and download link.