Arcane

Beta software — Arcane is in active development. APIs and storage formats may change between releases. Pin your version and back up ~/.arcane before upgrading.

Unified engineering intelligence — persistent memory, decision journeys, and knowledge capture for AI-assisted development workflows.

Arcane runs as an MCP (Model Context Protocol) server, giving Claude Code, Claude Desktop, and other MCP-compatible agents a persistent, searchable knowledge store that survives context window resets and spans every project you work on.

What It Does

• Memories — save and search decisions, bugs, patterns, and learnings with hybrid FTS + vector search
• Journeys — track multi-step investigations from problem → exploration → decision → outcome
• Artifacts — ingest CI runs, git commits, and Linear tickets as searchable references
• Relationships — link any entities (memory → memory, journey → artifact, etc.) into a knowledge graph
• Content generation — draft blog posts and Architecture Decision Records from your stored knowledge
• Intelligence — detect CI flake patterns and summarise engineering velocity

Quickstart

Install

# With uv (recommended)
uv tool install arcane

# Or with pip
pip install arcane

Initialise

arcane init

This creates ~/.arcane/ with a SQLite database and default config.

Connect to Claude Code

Add to your Claude Code MCP config (~/.claude/config.json or project .claude/config.json):

{
  "mcpServers": {
    "arcane": {
      "command": "arcane",
      "args": ["mcp"]
    }
  }
}

Restart Claude Code — Arcane tools will be available automatically.

Connect to Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "arcane": {
      "command": "arcane",
      "args": ["mcp"]
    }
  }
}

MCP Tools

Once connected, Claude has access to these tools:

CLI Reference

# Memory
arcane save                     # Interactive save
arcane search "query"           # Hybrid search
arcane context                  # Print context for agent injection
arcane details <id-prefix>      # Full memory details
arcane delete <id-prefix>       # Delete a memory
arcane reindex                  # Rebuild vector index
arcane stats                    # DB statistics
arcane sessions                 # List recent sessions

# Journeys
arcane journey start            # Start a journey
arcane journey update <id>      # Add an update
arcane journey complete <id>    # Mark complete
arcane journey list             # List journeys
arcane journey show <id>        # Full journey with linked entities

# Ingestion
arcane ingest git               # Ingest local git commits
arcane ingest gha               # Ingest GitHub Actions runs
arcane ingest linear            # Ingest Linear tickets

# Intelligence
arcane analyze flakes           # Detect CI flakes
arcane analyze velocity         # Engineering velocity summary

# Content
arcane draft blog               # Blog brief from memories
arcane draft adr <memory-id>    # ADR from a decision memory

# Relationships
arcane link <type-id> <type-id> <rel-type>   # Create link
arcane trace <type> <id>                      # Walk graph

# Config
arcane config                   # Show current config
arcane config set-home <path>   # Set custom data directory
arcane config clear-home        # Remove custom home setting

# Server
arcane mcp                      # Start MCP server (stdio)
arcane -v mcp                   # With debug logging

Configuration

Config is loaded from ~/.arcane/config.yaml (or $ARCANE_HOME/config.yaml):

embedding:
  provider: ollama           # "ollama" or "openai"
  model: nomic-embed-text    # Embedding model name
  base_url: http://localhost:11434  # Ollama base URL (ignored for openai)
  api_key: null              # OpenAI API key (or set OPENAI_API_KEY env var)

context:
  semantic: auto             # "auto" | "always" | "never"
  topup_recent: true         # Supplement semantic results with recent memories

Environment Variables

Semantic Search

Arcane supports two embedding backends:

Ollama (default, local, free)

# Install Ollama: https://ollama.ai
ollama pull nomic-embed-text

# Config (default — no changes needed)
embedding:
  provider: ollama
  model: nomic-embed-text

OpenAI

# Set API key
export OPENAI_API_KEY=sk-...

# config.yaml
embedding:
  provider: openai
  model: text-embedding-3-small

After switching models, rebuild the vector index:

arcane reindex

Memory Categories

Plugin System

Arcane uses Python entry points for extensibility. Install any package that declares the right entry point and Arcane will discover it automatically.

# In your plugin package's pyproject.toml
[project.entry-points."arcane.plugins.ingestion"]
jira = "my_package:JiraIngestionPlugin"

[project.entry-points."arcane.plugins.intelligence"]
code_churn = "my_package:CodeChurnAnalyser"

[project.entry-points."arcane.plugins.content"]
changelog = "my_package:ChangelogGenerator"

Plugins must implement the protocols defined in arcane.plugins.protocols.

Data Layout

~/.arcane/
├── arcane.db          # SQLite database (memories, journeys, artifacts, relationships)
└── vault/
    └── <project>/
        └── YYYY-MM-DD-session.md   # Markdown mirror of saved memories

All data lives in a single SQLite file — easy to back up, sync, or inspect with any SQLite tool.

Development

git clone https://github.com/dkelly/arcane
cd arcane
uv venv && source .venv/bin/activate
uv pip install -e ".[dev]"
pre-commit install

# Run tests
pytest

# Lint + format
ruff check --fix src/ tests/
ruff format src/ tests/

# Type check
mypy src/arcane

Project Structure

src/arcane/
├── cli/            # Click CLI — one module per command group
├── domain/         # Pydantic domain models
├── infra/          # DB repos, config, search, embeddings, redaction
├── mcp_server/     # MCP stdio server + tool handlers
├── plugins/        # Plugin protocols + built-in implementations
└── services/       # Business logic layer

See AGENTS.md for detailed contributor and agent guidance.

License

MIT