Grimoire AI

Local-first, AI-native requirements management for software projects.

Grimoire stores project knowledge — features, requirements, tasks, and architecture decisions — as structured markdown files in your git repository. A CLI designed for AI coding agents provides fast, structured access to project context.

Grimoire answers the question: "How does a new AI agent session get oriented in a project — fast?"

Quick start

# Install the CLI
npm install -g @grimoire-ai/cli

# Initialize grimoire in your project
grimoire init --name "My Project"

# Install AI agent skills (agentskills.io)
npx skills add mikevalstar/grimoire

Usage

Read project context

grimoire overview                          # Project overview
grimoire feature list                      # List all features
grimoire task list --status todo           # Find open work

Create documents

grimoire feature create --title "User Authentication" --priority high --tag security
grimoire requirement create --title "OAuth 2.0 Login" --feature feat-xxxxx-user-authentication
grimoire task create --title "Setup Google OAuth" --requirement req-xxxxx-oauth-20-login
grimoire decision create --title "Use JWT Over Sessions" --status accepted

Update and track progress

grimoire task update <id> --status in-progress
grimoire log <id> "Implemented OAuth callback handler" --author claude-code
grimoire comment <id> "Should we support SAML as well?"

Web UI

grimoire ui                                # Launch web dashboard on port 4444
grimoire ui --port 8080                    # Custom port

The web UI provides a visual dashboard with document browsing, filtering, sorting, and rendered markdown.

Validate

grimoire validate                          # Check for broken links, missing fields

Document types

Type	Directory	Purpose
overview	overview.md	Single project overview
feature	features/	High-level capabilities
requirement	requirements/	Detailed specs, linked to features
task	tasks/	Implementation work items
decision	decisions/	Architecture Decision Records (ADRs)

All documents are markdown files with YAML frontmatter, stored in .grimoire/ and committed to git.

How it works

.grimoire/
  overview.md          # Project overview
  config.yaml          # Configuration
  features/            # Feature documents
  requirements/        # Requirement documents
  tasks/               # Task documents
  decisions/           # Architecture decisions
  .cache/              # Gitignored — derived database

Markdown files are the source of truth. The database (DuckDB) is a derived cache that enables full-text search, semantic search, and relational queries — it's always rebuildable from files via grimoire sync.

AI agent workflow

Grimoire is designed for AI coding agents (Claude Code, Cursor, Copilot, etc.) to consume. All commands output structured JSON by default.

# Agent starting work
grimoire overview                                    # Understand the project
grimoire task list --status todo                     # Find available work
grimoire task get <id>                               # Read task details
grimoire task update <id> --status in-progress       # Claim a task

# Agent recording progress
grimoire log <id> "Completed implementation"         # Log what was done
grimoire decision create --title "..." --body "..."  # Record decisions
grimoire task update <id> --status done              # Mark complete

Install the agentskills.io skill to give your AI agent full knowledge of Grimoire commands:

npx skills add mikevalstar/grimoire

Output format

All commands output JSON by default (AI mode). Use --format cli for human-readable output:

grimoire feature list --format cli

Development

See DEVELOPMENT.md for setup, building, testing, and contributing.

License

MIT