What if you could git diff your plot? Laires treats your novel or screenplay like a codebase — it builds a narrative graph of characters, conflicts, and arcs from your manuscript, then gives you an LLM agent with 28 tools to query, lint, and co-edit your story.
Word processors know nothing about your story. They see paragraphs, not characters. Laires bridges the gap — it parses your manuscript into a structured model, then lets an AI agent reason about plot, consistency, and perspective using that structure. Think of it as IDE-level tooling for writing.
Alpha — functional and tested (225+ tests), but APIs may change. Feedback and contributions welcome.
- Narrative graph — Extracts characters, objectives, conflicts, and relationships into a structured, queryable graph
- Native desktop GUI — Single-binary desktop app with chat, canvas, scene sidebar, and force-directed graph visualization
- Two operating modes — Consultant for read-only analysis with revision briefs; Workshop for live co-editing
- Multi-turn agent chat — 28 built-in tools for querying the graph, searching text, running lint checks, analyzing pacing, and more
- Character perspectives — LLM-powered subjective interpretation of scenes through any character's eyes
- Multi-file projects — Novels, screenplays, or any multi-document project with automatic file discovery
- Prose and Fountain — Markdown prose and Fountain screenplay format, plus .docx import
- Multiple LLM providers — Anthropic, OpenAI, Gemini, Ollama, or any OpenAI-compatible endpoint
- Consistency checking — Lint rules and divergence detection to catch contradictions and plot holes
- Custom skills — Extend the agent with your own TOML-defined tools
- VCS integration —
laires diffandlaires logshow narrative graph changes across git/jj commits
- Rust toolchain (1.85+, edition 2024)
- An LLM API key (Gemini, Anthropic, OpenAI, or a local model via Ollama)
git clone https://github.com/justingibbs/laires.git && cd laires
cargo install --path .# Create a project
mkdir my-novel && cd my-novel
# Add your API key
echo 'GEMINI_API_KEY=your-key-here' > .env
# Initialize
laires init --title "My Novel"
# Write your story in story.md, then analyze it
laires scan
# Explore
laires graph # print narrative graph summary
laires chat # interactive agent chat
laires gui # launch the desktop GUISeparate scenes with Markdown headings (## Chapter Title) or horizontal rules (---). For screenplays, use laires init --title "Title" --fountain and standard Fountain scene headings (INT. COFFEE SHOP - DAY).
| Command | Description |
|---|---|
laires init --title "Title" |
Initialize a new project |
laires scan |
Analyze all scenes with the LLM |
laires graph |
Print narrative graph summary |
laires status |
Show project stats |
laires chat |
Interactive agent chat |
laires gui |
Native desktop GUI |
laires open |
Split-pane terminal UI with overlays |
laires lint |
Run consistency checks |
laires diff |
Graph changes since last commit |
laires perspective <char> |
View a scene through a character's eyes |
laires brief |
View or list revision briefs |
laires convert <file> |
Convert .docx/.txt to .md |
Run laires --help for full usage details.
The config file at .laires/config.toml is created by laires init:
[llm]
provider = "gemini"
model = "gemini-2.5-flash"
api_key_env = "GEMINI_API_KEY"
base_url = "https://generativelanguage.googleapis.com/v1beta/openai"
[project]
title = "My Novel"
format = "prose"
[analysis]
debounce_ms = 2000
auto_scan = true| Provider | provider value |
api_key_env |
base_url |
|---|---|---|---|
| Google Gemini | "gemini" |
GEMINI_API_KEY |
https://generativelanguage.googleapis.com/v1beta/openai |
| Anthropic | "anthropic" |
ANTHROPIC_API_KEY |
https://api.anthropic.com |
| OpenAI | "openai" |
OPENAI_API_KEY |
https://api.openai.com/v1 |
| Local (Ollama) | "local" |
— | http://localhost:11434/v1 |
Laires is built on the Concept & Synchronization pattern (Jackson & Meng, MIT CSAIL) — 10 independent concept modules coordinated through explicit synchronizations. See docs/overview.md for the full technical overview.
- No streaming responses yet — the agent returns full replies after processing
- Graph visualization in the GUI is functional but basic (no arc overlays)
- Edition 2024 requires a recent Rust toolchain (1.85+)
- LLM analysis quality depends on the model — larger models produce better narrative graphs
cargo build # build (debug)
cargo test # run tests (~225 tests)
cargo fmt # format code
cargo clippy # lintSee CONTRIBUTING.md for guidelines on submitting changes.