SwarmVault

SwarmVault is a local-first LLM knowledge compiler.

It ingests raw research inputs, stores them immutably, and compiles them into a persistent markdown wiki, a structured graph, and a local search index. Instead of losing work inside chat history, the vault becomes the durable artifact.

What You Get

  • Local-first artifacts - keep sources, wiki pages, graph data, and search state on disk
  • **Schema-guided compilation** - teach each vault its own naming, categorization, and grounding rules with swarmvault.schema.md
  • **Markdown + graph outputs** - generate readable wiki pages and structured graph.json
  • Pluggable providers - use OpenAI, Anthropic, Gemini, Ollama, generic OpenAI-compatible APIs, custom adapters, or the built-in heuristic
  • **Automation loop** - import from inbox/, watch for changes, and keep run logs in state/jobs.ndjson
  • **MCP access** - expose the vault to compatible agents with swarmvault mcp
  • Agent-ready - install rules for Codex, Claude, and Cursor

Install

SwarmVault requires Node >=24.

npm install -g @swarmvaultai/cli

Core Workflow

  1. Ingest a file path or URL into immutable source storage
  2. **Shape the vault** with swarmvault.schema.md
  3. Compile source manifests into wiki pages, graph data, and local search
  4. **Query** the compiled vault and save useful answers back into wiki/outputs/
  5. **Automate** capture workflows with inbox import, watch, and MCP

Quick Start

swarmvault init
swarmvault ingest ./my-document.pdf
swarmvault compile
swarmvault query "What are the key concepts?" --save
swarmvault graph serve

Why SwarmVault

Most "chat with your docs" tools answer a question and discard the work. SwarmVault keeps the work. Source manifests, schema rules, markdown pages, query outputs, graph edges, and freshness state all remain as inspectable files you can diff, search, and reuse.