LitAssist

Last Updated: October 12, 2025

LitAssist is a comprehensive legal workflow automation tool designed for Australian legal practice. It provides a structured end-to-end pipeline for litigation support:

plan → research → analyse → structure → brainstorm → strategy → draft

Start with CasePlan:
ALWAYS begin with the caseplan command, even with a skeleton case_facts.txt file. It generates a complete litigation roadmap with executable commands, helping you systematically build your case. The plan includes cost estimates, time projections, and explains every technical choice.

graph TD
    CP["CasePlan<br/>Start Here!"] --> A["Lookup<br/>Research"]
    CP --> B["Digest<br/>Analyse"]
    CP --> C["ExtractFacts<br/>Structure"]
    
    A --> D["Brainstorm<br/>Generate Options"]
    B --> D
    C --> D
    
    D --> E["Strategy<br/>Plan Approach"]
    E --> F["Draft<br/>Create Documents"]
    
    B --> CN["CounselNotes<br/>Strategic Analysis"]
    C --> CN
    CN --> E
    
    C --> BB["Barbrief<br/>Barrister's Brief"]
    D --> BB
    E --> BB
    A --> BB
    
    G[Utilities] --> H["Test<br/>API Check"]
    G --> I["Audit<br/>Logging"]
    G --> J["Mock<br/>Mode"]
    
    style CP fill:#f9f,stroke:#333,stroke-width:4px
    style A fill:#e1f5fe
    style B fill:#e1f5fe
    style C fill:#e1f5fe
    style D fill:#fff3e0
    style E fill:#fff3e0
    style F fill:#e8f5e9
    style CN fill:#f3e5f5
    style BB fill:#f3e5f5

Core Commands

Start Here: CasePlan

CasePlan: START HERE! Automated phased workflow planning with executable scripts
- Generates complete litigation roadmaps tailored to your case
- Creates executable bash scripts with all commands ready to run
- Explains every technical choice with "Switch rationale" comments
- Adapts to case complexity: minimal, standard, or comprehensive workflows

Primary Workflow Commands

Lookup: Rapid case-law research (Jade.io database via Google Custom Search + Gemini 2.5 Pro)
Digest: Mass document processing (chronological summaries or issue-spotting via Claude Sonnet 4.5)
ExtractFacts: Automatic extraction of case facts into a structured file (Claude Sonnet 4.5)
Brainstorm: Creative legal strategy generation (unorthodox strategies via Grok 4, analysis via o3-pro)
Strategy: Targeted legal options with probability assessments (state-of-the-art litigation reasoning via Claude Sonnet 4.5)
Draft: Citation-rich document creation (superior technical writing via o3-pro)

Specialized Commands

CounselNotes: Strategic advocate analysis with structured extractions (o3-pro)
Barbrief: Comprehensive barrister's briefs for litigation (extended output via o3-pro)

For detailed usage guidance, see the LitAssist User Guide.

Installation

Quick Start

# 1. Clone repository
git clone https://github.com/your-org/litassist.git
cd litassist

# 2. Install globally with pipx (recommended)
brew install pipx
pipx install -e .
pipx inject litassist tiktoken
pipx ensurepath
source ~/.zshrc

# For local development with pip instead of pipx
# use the pinned requirements to avoid version issues
# (openai==0.28.1, google-api-python-client, etc.)
#
# pip install -r requirements.txt

# 3. Setup configuration
cp config.yaml.template config.yaml
# Edit config.yaml with your API keys

# 4. Copy config to global location
mkdir -p ~/.config/litassist
cp config.yaml ~/.config/litassist/

# 5. Use from anywhere
cd ~/any-directory/
litassist caseplan case_facts.txt

For detailed installation options, troubleshooting, and advanced setup, see the Installation Guide

Configuration

Required API keys in config.yaml (see config.yaml.template for reference):

openrouter:
  api_key:    "YOUR_OPENROUTER_KEY"
  api_base:   "https://openrouter.ai/api/v1"   # optional

openai:
  api_key:          "YOUR_OPENAI_KEY"  # For embeddings and o3-pro/GPT-5 BYOK via OpenRouter
  embedding_model:  "text-embedding-3-small"

google_cse:
  api_key:                "YOUR_GOOGLE_API_KEY"         # API key for Google Custom Search
  cse_id:                 "YOUR_JADE_CSE_ID"           # Google CSE for Jade.io
  cse_id_comprehensive:   "YOUR_COMPREHENSIVE_CSE_ID"  # Optional: broader legal sources (gov.au etc.)
  cse_id_austlii:         "YOUR_AUSTLII_CSE_ID"        # Optional: AustLII CSE for Australian legal cases

pinecone:
  api_key:     "YOUR_PINECONE_KEY"
  environment: "YOUR_PINECONE_ENV"   # e.g. "us-east-1-aws"
  index_name:  "legal-rag"

llm:
  use_token_limits: false    # Enable conservative token limits for AI responses (default: false = use model defaults)

general:
  heartbeat_interval: 10  # Progress indicator interval in seconds (default: 10)
  max_chars: 200000       # Document chunking: characters per chunk for digest/extractfacts (default: 200000 ≈ 50K tokens)
  rag_max_chars: 8000     # Document chunking: characters per chunk for draft command embeddings (default: 8000 ≈ 1600 words)

OpenRouter Configuration

OpenRouter is the primary API gateway for all LLM calls. Some models require BYOK (Bring Your Own Key) setup:

Models requiring BYOK:

OpenAI o3-pro (draft, barbrief, counselnotes commands)
OpenAI GPT-5 and GPT-5 Pro (critical verification commands)
Claude Sonnet 4.5 available without BYOK

Quick BYOK Setup:

Go to OpenRouter Settings
Add your API keys under integrations:
- OpenAI: Requires Tier 4+ API key with o3-pro and GPT-5 access (standard keys won't work)
- Anthropic: Any valid API key (Claude models available without BYOK on OpenRouter)
Save and verify model availability on your dashboard

Model Configuration & BYOK Requirements

LitAssist uses a three-tier strategy with cutting-edge AI models optimized for legal work:

Three-Tier Model Strategy (October 2025):

Tier 1: Critical Verification - GPT-5 Pro (<1% hallucination rate) for soundness checking
Tier 2: Fast Verification - GPT-5 (1.4% hallucination rate) for standard verification
Tier 3: Legal Reasoning - Claude Sonnet 4.5 (state-of-the-art for complex litigation tasks)

Command	Model	BYOK Required	Purpose
caseplan	Claude Sonnet 4.5	No	Workflow planning - START HERE!
lookup	Google Gemini 2.5 Pro	No	Rapid case-law research with 1M context
digest	Claude Sonnet 4.5	No	Document analysis and issue identification
extractfacts	Claude Sonnet 4.5	No	Structured fact extraction with citations
brainstorm	Claude Sonnet 4.5 / Grok 4	No	Legal strategies + creative ideation
strategy	Claude Sonnet 4.5	No	State-of-the-art legal reasoning
draft	OpenAI o3-pro	Yes	Superior technical legal writing
counselnotes	OpenAI o3-pro	Yes	Strategic advocate analysis
barbrief	OpenAI o3-pro	Yes	Comprehensive barrister's briefs
verify (critical)	GPT-5 Pro	Yes	Critical soundness verification
verify (standard)	GPT-5	Yes	Fast verification with high accuracy

Why These Models?

Claude Sonnet 4.5 (September 2025):

Explicitly "state of the art on complex litigation tasks"
Superior legal domain knowledge and reasoning
80% cost reduction vs previous models
Extended thinking mode for multi-step analysis

GPT-5 and GPT-5 Pro (August 2025):

Industry-leading accuracy: <1.6% hallucination rate
6x fewer factual errors than previous models
Critical for legal verification where accuracy is paramount

OpenAI o3-pro:

Advanced reasoning for technical drafting
Extended output capacity (32K tokens) for briefs
Specialized for structured legal documents

Required BYOK Setup

Commands using restricted models (draft, barbrief, counselnotes, verification) require BYOK setup on OpenRouter. Without BYOK, these commands will fail with authentication errors.

Note: Check the OpenRouter Models page to verify which models are available with your API key vs. requiring BYOK.

For detailed API configuration, model availability, and troubleshooting, see the LitAssist User Guide.

Command Reference

Basic usage:

litassist [GLOBAL OPTIONS] <command> [ARGS] [OPTIONS]

Global options:

--log-format [json|markdown]: Choose audit-log format (default: json)
--verbose: Enable debug-level logging

Core Pipeline Commands

1. caseplan - Your Starting Point

ALWAYS START HERE! This command is your litigation GPS, generating a complete roadmap tailored to your case.

# Step 1: Get budget recommendation based on case complexity
litassist caseplan case_facts.txt

# Step 2: Generate full plan with executable script
litassist caseplan case_facts.txt --budget standard

# Step 3: Execute the generated workflow
bash caseplan_commands_standard_*.txt

Key Features:

Complexity Analysis: Evaluates legal, factual, procedural, and strategic complexity
Budget Levels:
- minimal: Quick matters, 5-7 phases
- standard: Typical litigation, 10-12 phases
- comprehensive: Complex cases, 15-25 phases
Context: Use --context "breach of contract" to guide analysis and prioritize relevant workflow phases
Executable Output: Creates bash script with all commands ready to run
Switch Explanations: Every technical choice explained inline

Example Output:

# Phase 1: Extract Initial Facts
litassist extractfacts contract.pdf emails.pdf
# Switch rationale: Standard extraction, no special parameters needed

# Phase 2: Research Contract Breach Elements
litassist lookup "elements of contract breach and damages" --mode irac --comprehensive
# Switch rationale: --comprehensive for detailed analysis, --mode irac for structured output

2. lookup - Rapid case-law search with automatic citation

litassist lookup "What defences exist to adverse costs orders?"
litassist lookup "Question?" --mode broad --comprehensive
litassist lookup "contract formation elements" --extract citations
litassist lookup "negligence principles" --extract principles  
litassist lookup "discovery requirements" --extract checklist

Options:

--mode [irac|broad]: Analysis format (default: irac)
--extract [citations|principles|checklist]: Extract specific information in structured format
--comprehensive: Enable comprehensive mode: standard searches yield up to 5 results each from Jade and AustLII; comprehensive mode yields up to 10 results each from Jade, AustLII, and a secondary CSE.

3. digest - Process large documents for summaries or issues

litassist digest bundle.pdf --mode [summary|issues]

4. extractfacts - Extract structured case facts from documents

# Single file
litassist extractfacts document.pdf

# Multiple files
litassist extractfacts file1.pdf file2.txt file3.pdf

# Creates: extractfacts_[combined_slugs]_YYYYMMDD_HHMMSS.txt
# Note: case_facts.txt must be created or edited manually

5. brainstorm - Generate comprehensive legal strategies with reasoning traces

# Default: uses case_facts.txt if present in current directory
litassist brainstorm --side [plaintiff|defendant|accused] --area [criminal|civil|family|commercial|administrative]

# Specify facts file(s) explicitly
litassist brainstorm --facts case_facts.txt --side plaintiff --area civil

# Use multiple facts files with glob patterns
litassist brainstorm --facts 'case_*.txt' --side plaintiff --area civil

# Add research context (supports glob patterns)
litassist brainstorm --side plaintiff --area civil --research 'outputs/lookup_*.txt'

# Multiple research files with selective patterns
litassist brainstorm --side plaintiff --area civil --research 'outputs/lookup_*gift*.txt' --research 'outputs/lookup_*trust*.txt'

# Creates: brainstorm_[area]_[side]_YYYYMMDD_HHMMSS.txt (main strategies)
#          brainstorm_[area]_[side]_YYYYMMDD_HHMMSS_orthodox_reasoning.txt
#          brainstorm_[area]_[side]_YYYYMMDD_HHMMSS_unorthodox_reasoning.txt  
#          brainstorm_[area]_[side]_YYYYMMDD_HHMMSS_analysis_reasoning.txt
# Note: strategies.txt must be created or edited manually

6. strategy - Generate targeted legal options and draft documents

litassist strategy case_facts.txt --outcome "Obtain interim injunction against defendant"
# Or incorporate brainstormed strategies
litassist strategy case_facts.txt --outcome "..." --strategies strategies.txt

7. draft - Create citation-rich legal drafts with intelligent document recognition

# Single document
litassist draft case_facts.txt "skeleton argument on jurisdictional error"
# Multiple documents (automatically recognizes case_facts.txt and strategies.txt)
litassist draft case_facts.txt strategies.txt "argument based on strategy #3"
# Mix text files and PDFs
litassist draft case_facts.txt bundle.pdf "comprehensive submission"

8. counselnotes - Generate strategic advocate analysis

# Single document
litassist counselnotes document.pdf
# Multiple documents
litassist counselnotes doc1.pdf doc2.txt doc3.pdf
# Extract specific sections
litassist counselnotes case_bundle.pdf --extract all

9. barbrief - Generate comprehensive barrister's briefs for litigation

# Basic brief for trial
litassist barbrief case_facts.txt --hearing-type trial
# Appeal brief with strategies
litassist barbrief case_facts.txt --hearing-type appeal --strategies strategies.txt
# Full brief with all materials
litassist barbrief case_facts.txt --hearing-type interlocutory \
  --strategies strategies.txt \
  --research lookup_report1.txt --research lookup_report2.txt \
  --documents affidavit.pdf --documents exhibit_a.pdf \
  --context "Focus on jurisdictional issues" \
  --verify

Required:

Case facts in 10-heading format (from extractfacts)
--hearing-type [trial|directions|interlocutory|appeal]

Options:

--strategies FILE: Brainstormed strategies
--research FILE: Lookup/research reports (multiple allowed)
--documents FILE: Supporting documents (multiple allowed)
--context TEXT: Additional context to guide the analysis
--verify: Enable citation verification

Utility Commands

test - Verify API connectivity
```
litassist test
```

Example Files

The examples/ directory contains sample files for testing all commands, based on the fictional Smith v Jones family law case.

Output Files & Logging

Command Output Files

All commands now save their output to timestamped text files without overwriting existing files:

caseplan: caseplan_[budget]_YYYYMMDD_HHMMSS.txt and caseplan_commands_[budget]_YYYYMMDD_HHMMSS.txt
lookup: lookup_[query_slug]_YYYYMMDD_HHMMSS.txt
digest: digest_[mode]_[filename_slug]_YYYYMMDD_HHMMSS.txt
brainstorm: brainstorm_[area]_[side]_YYYYMMDD_HHMMSS.txt
extractfacts: extractfacts_[filename_slug]_YYYYMMDD_HHMMSS.txt
strategy: strategy_[outcome_slug]_YYYYMMDD_HHMMSS.txt
draft: draft_[query_slug]_YYYYMMDD_HHMMSS.txt
counselnotes: counselnotes_[filename_slug]_YYYYMMDD_HHMMSS.txt
barbrief: barbrief_[hearing_type]_YYYYMMDD_HHMMSS.txt

Each output file includes metadata headers with command parameters and timestamps.

Output Organization

Command outputs are automatically stored in the outputs/ directory
Detailed logs are saved in logs/<command>_YYYYMMDD-HHMMSS.{json|md}
Progress indicators keep you informed during long-running operations (configurable heartbeat interval)
Network errors are caught and displayed with user-friendly messages

Model Configuration

Each command uses optimized LLM models and parameters:

Factual tasks (lookup, extractfacts): temperature=0 for accuracy
Creative tasks (brainstorm, draft): temperature=0.5-0.9 for innovation
Verification: Always uses temperature=0 for consistency

Note:

Document chunking (max_chars) and AI output limits (use_token_limits) are separate systems.
tiktoken is required for accurate token counting and large document handling.
All output is now ASCII/ANSI only (no emoji).
See the LitAssist User Guide for details.

Disclaimer

This tool provides drafts and summaries only. All outputs must be reviewed by qualified legal counsel before filing or submission.

For detailed instructions, workflows, and examples, see the LitAssist User Guide.

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
article		article
demo		demo
docs		docs
examples		examples
litassist		litassist
test-scripts		test-scripts
tests		tests
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
INSTALLATION.md		INSTALLATION.md
MANIFEST.in		MANIFEST.in
README.md		README.md
RELEASE_PROCESS.md		RELEASE_PROCESS.md
TODO.md		TODO.md
cleanup.zsh		cleanup.zsh
config.yaml.template		config.yaml.template
litassist.py		litassist.py
pytest.ini		pytest.ini
requirements-ci.txt		requirements-ci.txt
requirements-test.txt		requirements-test.txt
requirements.txt		requirements.txt
setup.py		setup.py

agelastic/litassist

Folders and files

Latest commit

History

Repository files navigation