What is Indexed

Indexed is a local-first indexing engine that gives AI agents deep context over your codebase, documentation, Jira tickets, and Confluence pages. It builds a hybrid search index (BM25 + dense vectors) on your machine and exposes it via the Model Context Protocol (MCP) — so any MCP-compatible AI tool can search your indexed knowledge as naturally as it reads files.

Works with Claude Code, Claude Desktop, OpenAI Codex, Cursor, GitHub Copilot, and any other MCP-compatible agent.

What

Indexed turns your local sources into searchable collections. A collection is a named, typed snapshot of a source — files, a Jira project, or a Confluence space — stored as a local FAISS vector index at ~/.indexed/data/collections/<name>/.

~/.indexed/data/collections/my-docs/
├── manifest.json      # source path, connector type, timestamps
├── documents.json     # parsed document metadata
├── chunks.json        # chunk text and positions
└── index.faiss        # the binary vector index

Three steps to working search:

1. Connect — point Indexed at a source (a directory, a Jira project, or a Confluence space)
2. Index — code is parsed via tree-sitter AST, documents via Docling, everything is chunked and embedded using a local model. No API calls, no data sent anywhere.
3. Search — hybrid BM25 + dense vector retrieval from the CLI, or let AI agents search via MCP

Why

Your Jira tickets are a black hole

Your team has two years of Jira tickets — decisions, bug investigations, architecture discussions, deployment runbooks. A new engineer asks "how do we handle auth token rotation?" The answer is in PROJ-1847 from eight months ago, but Jira search is keyword-based and the ticket title says "SSO refresh logic." Indexed makes all of it semantically searchable — from the CLI or directly through your AI agent.

Your AI agent can't see your full codebase

Coding agents like Claude Code read files one at a time and grep for keywords. That works for small repos, but it burns tokens on large ones and misses semantic connections. Indexed gives agents a pre-built search index over your entire codebase — AST-aware via tree-sitter — so they find the right function, module, or pattern across thousands of files in milliseconds.

Your docs and wikis are invisible to AI

Confluence pages, Markdown vaults, Obsidian notebooks, Notion exports, PDFs — your team wrote all of it, but your AI assistant can't see any of it. Indexed parses and indexes them with Docling (including OCR for scanned documents) and makes them searchable through one unified MCP endpoint.

Privacy

What stays local — everything:

• Parsing, chunking, and embedding run entirely on your machine
• The FAISS vector index lives at ~/.indexed/ and never leaves your disk
• Search queries are embedded locally; no query text is sent to any external service

What network calls happen:

After the initial model download, Indexed works completely offline. There is no telemetry, no analytics, no usage reporting of any kind.

Continue reading: Quick Start · Indexing Overview · Configuration Guide