← All Deployments
Live — Active Federal Case
Litigation & Legal AI

PleaBrain

Not a document chatbot. A case-building machine. It builds your timeline. It surfaces what you don't know yet. It constructs evidence-backed narratives to win arguments. It writes proper legal motions — and waits for your attorney to approve them. Currently running inside an active $11.5 million ASBCA federal procurement dispute, with hundreds of attorney and paralegal hours already recaptured.

$11.5M
Active ASBCA federal case — real stakes, not a demo
Motions
Written, sourced, and submitted — with attorney approval
$250–
$800/hr — the billing rate behind every hour saved
100s
Of attorney/paralegal hours recaptured per case
The Timeline System

One prompt. A complete case architecture.

Drop 2,400+ documents. Ask for a timeline. Watch a full case chronology build itself — phases, impact scores, phase narratives, entity filters, source citations — all from a single instruction.

Phases Impact Scoring Narratives + Citations Swimlanes Threads Heatmap WHO / WHERE Filters Live Refresh Human Override
PleaBrain — Case Intelligence
You
Build a full timeline of this dispute. Rate each event's legal impact 0–10 and identify the case phases. Include source citations for every entry.
PleaBrain
Scanning 2,400 documents across Rule 4 record…
Extracting dated events and filing sequence…
Scoring legal impact for 24 entries…
Assigning phases + thread labels…
Generating phase narratives with citations…
Timeline built — 24 events · 3 phases · 9 key events →
Case Timeline
Phases
Scatter
Swimlanes
Heatmap
Narrative
Phase 1 — Contract Award
Source: Exhibit 001
8/10
KEY
Source: NTP Letter
5/10
Phase 2 — Performance Dispute
Source: Mod. Request #3
9/10
KEY
Source: CO Decision Letter
9/10
KEY
Source: Finance Memo Feb 2020
10/10
KEY
Phase 3 — ASBCA Claim
Source: Claim Document
9/10
KEY
Source: ASBCA Docket
7/10
KEY
AI Narrative — Phase 2: Performance Dispute
Phase 2 contains 3 key events. The CO's rejection of Modification Request #3 on Nov 14 is the pivotal break in the record — it precedes payment withholding by 3 months and anchors the delay attribution argument. Cross-reference the Nov–Dec email chain for the contracting officer's reasoning.
CO Decision Letter, Nov 2019 Finance Memo, Feb 2020 Email chain Nov 14–Dec 2
Impact Scoring — 0 to 10
Every event rated for legal weight. Color-coded nodes (purple → gold) visualize case strength at a glance. Key events (7+) are flagged and tracked separately. Scores can be set by the AI or overridden by the attorney.
Phase Narratives with Citations
Each phase gets an AI-generated narrative: the argument, the supporting evidence, the complications, and inline citation links back to the exact exhibits. Each phase narrative is a working section of a brief — ready for attorney review.
Swimlane View — Who Did What
Events organized by actor — one lane per party. See exactly what the contracting officer sent vs. what the contractor sent, in chronological order, in the same view. Deposition prep just became a 20-minute exercise.
Thread Cross-Referencing
Tag parallel storylines that span phases — "Payment dispute," "Scope creep," "Termination events." A thread view surfaces every event in that storyline across the full case timeline, regardless of phase assignment.
Entity Filters — WHO & WHERE
Names, organizations, and locations are extracted automatically from every evidence document. Filter chips let you view just the events involving a specific party or location — no manual tagging required.
Human Override at Every Layer
Every timeline entry, date, impact score, phase assignment, and narrative section can be reviewed, corrected, and overridden by the attorney. The AI does the structure. The attorney controls what goes in the record.

The Difference: Answering Questions vs. Building a Case

Most "AI for legal documents" products do one thing: you ask a question, it searches your documents and returns an answer. That's useful. It's also not remotely the full picture of what a legal team needs to win.

Winning a case requires more than finding documents. It requires constructing a narrative — a coherent story about what happened, in what order, who knew what when, and what the evidence proves. It requires identifying the gaps — the questions still open, the documents that should exist and don't, the claims that need more support. It requires building and maintaining a living timeline that connects events, communications, and exhibits into a structured picture of the dispute that gets sharper as more evidence is added.

The design principle: PleaBrain is not a lawyer and doesn't pretend to be. It's a tool that lets the legal team do things that were previously impossible — or that used to take paralegals weeks — in minutes. The human makes the legal judgments. The system handles the intelligence infrastructure.

What PleaBrain Actually Does

01

Active Case Timeline — Seven Views, Compounding Intelligence

Not a list of documents sorted by date. A structured case architecture built from the evidence — with AI reasoning layered on top. Every entry is linked to a specific source document. Every event is impact-scored 0–10 (colored purple → gold by legal weight). Every phase has an AI-generated narrative with inline citations back to the exact exhibits.

The timeline renders in seven views: the main scatter chart (event impact over time), chronological list, phase grid (events organized by legal phase), swimlane view (who did what, when — by actor), activity heatmap (density visualization), thread view (cross-cutting storylines), and narrative view (the full written case chronology with citations). One dataset, seven analytical angles.

Phase grouping is where it gets powerful. The AI identifies named phases from the evidence (e.g., "Contract Award," "Performance Dispute," "Claim Filing") and groups events accordingly — then generates a phase narrative for each one: the argument, the supporting evidence, the counter-evidence, and the sourced citations. Each phase narrative is a working section of a brief.

Thread cross-referencing lets you tag a parallel storyline (e.g., "Payment dispute," "Scope creep") that spans multiple phases. A thread view surfaces every event in that storyline across the full case timeline, regardless of which phase it falls in.

Entity-based filters (WHO, WHERE) let you view just the events involving a specific party or location — extracted automatically from the evidence, not manually tagged. The timeline refreshes live as new evidence is added. Human override is available for any entry, date, impact score, or narrative section.

7 views · Impact scoring 0–10 · Phase narratives + citations · Thread cross-referencing · WHO/WHERE entity filters · Timeline brief generation · Live refresh · Human override
02

Open-Loop Question Engine — Finding What You Don't Know

This is the piece that moves PleaBrain from reactive to proactive. As the system ingests and reasons over the corpus, it identifies and tracks open-loop questions — things the evidence raises but doesn't yet answer. A communication chain that ends abruptly. A claim made in one document that contradicts another. A meeting referenced in emails that has no corresponding notes in the record.

The system surfaces these gaps explicitly, prioritized by their potential impact on the case arguments being constructed. You learn what you still need to find — before opposing counsel makes that gap a liability.

Gap detection · Contradiction surfacing · Evidence completeness scoring
03

Evidence-Backed Narrative Construction

Not "here is a summary of your documents." Here is the story of what happened, told through the evidence, structured to support a specific argument. Ask PleaBrain to construct the narrative around a disputed claim and it returns a structured output: the argument, the evidence that supports it, the evidence that complicates it, and the source citations for every piece.

"Prove that the delay was attributable to material shortage, not contractor negligence." The system doesn't just retrieve documents that mention "delay" — it builds the argument, orders the supporting evidence, surfaces the counter-evidence, and produces a structured record you can hand to your attorney.

Argument structuring · Evidence ranking · Counter-evidence surfacing · Citation architecture
04

Compounding Case Intelligence

Every structured output builds on the last. The timeline becomes the context for narrative construction. The narratives surface new open-loop questions. The answers to those questions refine the timeline. The more the system structures, the more context it has — and the more sophisticated the reasoning it can do over the next query.

This is the compounding effect of structured output applied to a legal case: the system's intelligence about the dispute grows with every interaction, every ingested document, every structured record produced. A case worked through PleaBrain for three months has a fundamentally richer intelligence layer than one worked for three days.

Persistent fact store · Rolling case context · Cross-document inference · Session memory
05

Motion Drafting — Human in the Loop, AI Doing the Work

PleaBrain has drafted proper, procedurally correct legal motions — and those motions have been submitted. Not AI-generated boilerplate that an attorney discards. Structured drafts built from the actual case record: the specific timeline entries, the cited exhibits, the factual narrative the team has already constructed through the system.

The workflow is deliberately human-gated. The system produces a complete draft with every factual claim sourced to a specific document in the corpus. The attorney reviews, edits, and approves. Nothing goes out without sign-off. But the hours spent going from "blank page + a pile of documents" to "draft ready for attorney review" — those hours are now minutes. At $250–$500/hr for an associate or $400–$800/hr for a partner, that math is significant by the second motion.

This also means junior staff and paralegals can produce attorney-ready draft work product at a level that previously required much more senior time. The ceiling on what a paralegal can hand an attorney has moved substantially.

Motion drafting · Structured factual record · Source-anchored claims · Human approval gate · Iterative revision
06

Hybrid Retrieval Engine — Finds What Others Miss

The retrieval layer is why PleaBrain doesn't hallucinate answers. Every document is indexed two ways simultaneously: MeiliSearch for BM25 keyword search and PostgreSQL with pgvector for semantic vector similarity. Both indexes are queried in parallel and merged using Reciprocal Rank Fusion (RRF) — a rank aggregation algorithm that rewards documents appearing high in both sets. The result: you find the exhibit with the specific contract number AND the emails that semantically discuss the same dispute, even if they never use those exact words.

MeiliSearch BM25 · pgvector semantic · OpenAI embeddings · RRF fusion · Optional reranking
07

Initiative Agent Loop — Multi-Step Reasoning

Complex legal questions don't resolve in a single retrieval pass. The agent loop decomposes questions into sub-queries (facets), fires multiple searches, evaluates what came back, decides whether to fetch more, and only synthesizes when it has sufficient evidence. Built-in stall detection prevents circular retrieval loops. Every step is instrumented and returned in the response — so you can see exactly how the answer was built, not just what it says.

Custom agent loop · Facet tracking · Stall detection · GPT-4o-mini planner · GPT-4o synthesis
08

MCP Interface — All of Claude and GPT's Tools, Over Your Case

The entire platform is exposed as a Model Context Protocol (MCP) endpoint. This means Claude, ChatGPT with custom actions, or any AI tool can call directly into PleaBrain's case infrastructure — the timeline, the document corpus, the narrative engine, the open-loop questions — using the full reasoning power of those models. You get access to every capability of the frontier models, grounded in your specific case data, returning structured results you can act on.

MCP · OpenAPI Actions schema · Intent/scope/op routing · Async job handling

The Billing Rate Argument

Legal work is billed by the hour — and most of those hours aren't strategy. They're research, document review, drafting, and reorganizing what was found last week into what's needed this week. Those are exactly the hours PleaBrain eliminates or compresses dramatically.

Conservative estimate on a mid-size federal case: 300–500 hours of attorney and paralegal time across research, document review, motion drafting, deposition prep, and timeline reconstruction. At a blended rate of $200–$400/hr, that's $60,000–$200,000 in legal billing. PleaBrain doesn't eliminate attorney judgment — it eliminates the hours spent on work that isn't judgment.

Research: 3–4 hrs → <5 min

A complex factual research task — "what do our documents say about X event" — that required a paralegal half a day now returns a sourced, structured answer in minutes. Multiply by 50+ queries per case.

Motion Drafting: 8–16 hrs → 2–3 hrs

A motion that previously required an associate pulling documents, building the factual record, and drafting from scratch. Now: PleaBrain drafts with every claim sourced. The associate edits, tightens, approves. At $350/hr, saving 10 hours per motion is $3,500 — per motion.

Deposition Prep: Days → Hours

Building a comprehensive factual record for a deposition — every document the witness touched, every email they sent, the full timeline of their involvement — was days of manual work. Now it's a structured output built in an afternoon.

Paralegal Output Elevated

Junior staff can now produce attorney-ready work product that previously required senior associate time. The ceiling on what a paralegal can hand up the chain has moved substantially — which means more senior time goes to strategy, not review.

The human approval model is not a limitation — it's the design. Every motion, every narrative, every timeline entry can be reviewed, edited, and approved by the attorney before it matters. PleaBrain doesn't replace legal judgment. It removes the hours of labor that precede it, so the attorney's actual judgment — the thing that can't be automated — gets more time and better inputs.

The Timeline System: Bigger Than Litigation

The case timeline is the most underappreciated capability in PleaBrain — and it's the one that extends furthest beyond legal work. A structured, AI-maintained timeline that links every entry to its source documents and grows with new evidence is valuable in any domain where sequence matters:

Litigation & Legal Disputes

The canonical use case. Federal procurement, contract disputes, employment claims — any dispute where establishing the sequence of events is the work.

Insurance Claims Reconstruction

Property damage, liability, medical claims — reconstruct the event timeline from reports, photos, correspondence, and adjustor notes. Surface gaps before they become denials.

Construction & Project Disputes

RFIs, change orders, site logs, emails — timeline the project and prove (or defend) delay claims with sourced, structured evidence chains.

Regulatory & Compliance Investigation

When was the policy last updated? Who was notified of the change? What happened between the audit finding and the remediation? Timeline it — completely, from the documents.

What "Running in a Federal Case" Actually Means

The system is running in an active ASBCA federal procurement dispute with $11.5 million at stake. Not a demo. Not a test environment. A real case where the structured outputs produced by PleaBrain are being used by the legal team to build arguments, draft motions, identify evidence, and track the case's evolving narrative — where being wrong has real consequences.

That means the hallucination safeguards are real requirements, not best practices. The citation architecture isn't a feature — it's the entire point. The timeline isn't a nice-to-have — it's the working document the team lives in. When $11.5 million turns on what the record shows, "probably right" isn't good enough.

Hundreds of Billable Hours Recaptured

Research, document review, motion drafting, deposition prep — tasks that consumed hundreds of hours of attorney and paralegal time now take a fraction. Conservative estimate: $60,000–$200,000 in recaptured billing capacity per case.

Proper Motions. Written and Submitted.

Not boilerplate. Procedurally correct motions built from the actual case record — every factual claim sourced to a specific exhibit, reviewed and approved by the attorney before anything goes out.

Zero Unsourced Claims

No answer, no motion, no narrative surfaces without a document citation. Every claim traces back to a specific exhibit, email, or filing. Verifiable by opposing counsel.

Private by Design

Self-hosted. No case documents leave the client's infrastructure. Deployed behind HTTPS with Caddy. Opposing counsel never sees your corpus, your timeline, or your strategy.

Why This Is Hard to Replicate

Most "chat with your documents" products are thin wrappers around a vector database. Ask a question, get a paragraph. That's it. They fail at exactly the conditions that matter most in serious legal work:

  • Questions that require synthesizing across dozens of documents — a single retrieval pass isn't enough
  • Answers that need to be verifiable, not just plausible — citation architecture is a first-class requirement
  • Evidence gaps that need to be surfaced proactively — the system needs to know what it doesn't know
  • Case narrative that needs to hold together across months — not just a one-shot Q&A session
  • A living timeline that integrates new evidence without breaking existing structure

Every piece of PleaBrain was built to handle those conditions. The agent loop, the dual-index fusion, the timeline CRUD, the open-loop question engine, the compounding memory system — none of that exists in a simple RAG wrapper. That's why it's running in a federal case instead of a demo.

Deploy This in Your Domain

The core architecture — hybrid retrieval, agent loop, timeline system, narrative construction, open-loop question engine — is domain-agnostic. The same infrastructure that builds a federal procurement case builds an insurance claim reconstruction, a construction project dispute record, a regulatory compliance investigation, or any domain where winning requires controlling the narrative over a large evidence corpus.

The core infrastructure is proven in production. A new vertical deployment starts with a scoping call and moves to a working pilot in weeks.