⚖️

IRAC Maker

AI-powered legal writing practice for law school students

Eleven study modes across three categories. Runs entirely on your machine via Ollama. Optional BYOK for cloud models like Grok. Your facts, drafts, and uploaded outlines never leave your computer unless you explicitly opt in.

Get Started on Codeberg → See how it works
Eleven study modes

What you can do

Pick an area of law (Contracts, Torts, ConLaw, Crim, BizAssoc, ...). Then choose what kind of work you want to do.

Generate

— the AI does the analysis for you

Paste a fact pattern or court opinion → get a structured response.

  • IRACIssue, Rule (split into Statement + Exploration), Application, Conclusion + study tips. Streaming output.
  • Case BriefFacts, Procedural Posture, Issue, Holding, Reasoning, Dissent, exam notes.
  • Both SidesPlaintiff's strongest IRAC and Defendant's strongest IRAC, generated sequentially.

Practice

— you do the work, the AI grades or guides

Drills, multiple choice, essay grading, Socratic dialogue. Most modes include a ⚡ Generate hypo button so you don't need to bring your own fact pattern — pick area, pick topic (Agency, Negligence, 4th Amendment, etc.), and the LLM writes a fresh MEE-style hypo.

  • Issue SpottingList every issue you spot; AI scores coverage (caught / missed / false alarm).
  • MBE PracticeOne MBE-style question on demand with full per-choice explanations. Running session score.
  • Compare & FeedbackWrite your IRAC (sectioned or pasted whole); graded against the AI's model answer.
  • Long EssayBar-exam-length multi-issue essay. Per-issue breakdown + overall grade + coverage note.
  • Socratic ModeProfessor asks one question at a time. No answers given upfront — you spot the issues.

Library

— your stuff and settings

Manage outlines, browse history, switch LLM provider.

  • My OutlinesUpload your own (PDF/DOCX/TXT/MD) or use AI-generated, cached "built-in" rule references per area. 3-way pill: Mine / Built-in / None.
  • HistoryEvery IRAC, brief, drill, and essay auto-saved locally. Search, reopen, delete.
  • Settings (BYOK)Optional. Switch from local Ollama to a cloud LLM via your own API key. Currently supports Grok (xAI). Default stays local.
  • AboutMode reference and IRAC writing tips.
The framework

Why IRREAC, not IRAC?

Standard IRAC compresses the whole rule into one step. IRREAC splits it — and the split is worth points.

I

I — Issue

Frame the precise legal question the court must answer.
Whether [X] given [key facts].
R1

R1 — Rule Statement

State the applicable rule with a specific citation — Restatement §, UCC §, or landmark case.
Citation is mandatory. "The law requires…" earns no credit.
R2

R2 — Rule Exploration

Key upgrade
How have courts interpreted the rule? Key cases, majority vs minority, circuit splits.
This is what separates a B answer from an A.
A

A — Application

Break the rule into elements. Analyze each separately against the specific facts. Include counter-arguments.
~50% of your exam score lives here.
C

C — Conclusion

Answer the Issue directly. Add a confidence level (High / Moderate / Low) with a one-sentence reason.
Never introduce new analysis here.
💡 The two-Rule split (R1 + R2) is the key upgrade. Rule Exploration — how courts have interpreted the rule, key cases, majority vs minority views — is what separates a B answer from an A.
Get running

Two steps to start

1

Install Ollama

Download and run Ollama — it serves the local LLM that powers all analysis. Single drag-and-drop on macOS, single binary on Linux/Windows.

2

Clone and run

git clone https://codeberg.org/russkysong/IRAC-Maker.git
cd IRAC-Maker
bash setup.sh
streamlit run app.py

First run downloads the ~5.6 GB qwen3.5:9b model and bundles it into a custom irac-maker Ollama model with the IRREAC system prompt. Subsequent launches are instant.

Local-first by default

Privacy

🔒 Your work stays on your computer

Out of the box, every fact pattern, IRAC, case brief, uploaded outline, and saved history entry lives only on your machine. Nothing is uploaded, nothing is sent to a server, nothing is committed to git. The Ollama model runs locally; the app is single-user; the storage is plain JSON files under ~/.iracmaker/.

BYOK is opt-in. If you switch to a cloud model (Grok, GPT-4, etc.) in Settings, the app shows a red privacy notice and auto-flips the outline source away from your uploads to keep purchased commercial outlines off the network.

Outlines you upload are your responsibility. The app does not bundle any commercial outlines. If you upload copyrighted materials (Glannon, Emanuel, etc.), they're stored locally for your personal study use only.