Indexa is a free, local-first macOS app for building private AI-powered knowledge bases. Ask questions across your documents, share encrypted bundles, and connect to any AI tool — with no data ever leaving your computer.
Requires Ollama · Apple Silicon or Intel · No account needed
Download Ollama (free, open source) and pull an embedding and chat model. Indexa auto-connects on launch — no configuration needed.
Drag in files or paste URLs. Indexa ingests 15+ formats, crawls websites, and uses hybrid search for the most accurate retrieval.
Ask natural language questions in multi-turn conversations. Get streaming AI answers with source citations and relevance scores.
Use the MCP server with Claude Desktop or Cursor, query via OpenAI-compatible API, or export encrypted .indexa bundles.
Indexa ships with a built-in MCP (Model Context Protocol) server and a fully OpenAI-compatible API. Connect Claude Desktop, LM Studio, Cursor, or any OpenAI client directly to your local knowledge bases.
SSE + HTTP transport with three tools: query, search, and list_collections. Works with stdio clients via mcp-remote bridge.
Drop-in replacement for OpenAI clients. Collections appear as models. Supports /v1/chat/completions and /v1/models — no code changes needed.
BM25 keyword search + semantic vector search, merged via Reciprocal Rank Fusion. Optional query decomposition and LLM reranking for complex questions.
Full conversation history with context. Cross-collection search lets you query all your knowledge bases at once in a single question.
// Add Indexa to Claude Desktop { "mcpServers": { "indexa": { "command": "npx", "args": [ "mcp-remote", "http://localhost:11435/mcp" ] } } }
from openai import OpenAI client = OpenAI( base_url="http://localhost:11435/v1", api_key="your-api-key" ) response = client.chat.completions.create( model="Research Notes", # your collection messages=[{ "role": "user", "content": "Summarize the key findings" }] )
PDF, DOCX, TXT, Markdown, RTF, HTML, XLSX, PPTX, JSON, XML, YAML, CSV, log files, and source code. Drag and drop directly into the app.
Paste a URL to index a single page or crawl an entire site with configurable depth and page limits. SHA-256 change detection keeps content fresh automatically.
BM25 keyword + semantic vector search merged via Reciprocal Rank Fusion. Optional query decomposition and LLM reranking for complex, multi-part questions.
New in v1Organize documents into named collections with custom system prompts, refresh intervals, and query context. Cross-collection search available.
Password-protect in two modes: Sources Masked — answers only, no citations — or Read Only — view and query, no modifications. PBKDF2 with timing-safe comparison.
AI rewrites documents for better retrieval accuracy while preserving all facts, names, and technical details. Checkpoint recovery means long distillations survive crashes.
Export as portable .indexa bundles with AES-256-GCM encryption and PBKDF2 key derivation. Recipients query your knowledge — never your sources.
Ollama by default, or any OpenAI-compatible server. Use separate providers for chat and embeddings. Per-provider API keys and connection testing.
Web sources re-index on a schedule — hourly, daily, or weekly. SHA-256 content hashing means only changed content triggers a re-ingest.
Indexa runs a local HTTP server on port 11435. It speaks native REST and OpenAI-compatible endpoints — so existing tooling just works, no code changes required.
API key authentication, CORS restricted to localhost, and rate limiting on password attempts (10 tries, 5-minute lockout). Protection-aware — masked collections strip sources from all API responses automatically.
All document processing, indexing, and querying happens locally via Ollama. Nothing is sent to any cloud service.
Download, install, and use. No sign-up, no external API keys, no subscription.
Indexa collects no usage data. Your documents, queries, and knowledge bases are yours alone.
After a one-time model download via Ollama, Indexa requires no internet connection whatsoever.
Indexa is free for individuals. For organizations that need server-managed access control, user provisioning, and audit logging — enterprise features are on the roadmap. If that's you, we'd like to hear from you.
Knowledge base access controlled by your server. Revoke an associate's access instantly — the bundle stops working on their machine immediately.
Provision and deprovision users centrally. Define who can access which knowledge base without touching the bundle files.
See who queried what and when. Essential for compliance in legal, healthcare, and financial environments.
Run the access control server on your own infrastructure. Consistent with Indexa's privacy-first design — no data leaves your environment.
We're building enterprise features based on real demand. If your team needs server-managed access control, reach out — we'd love to learn about your use case.
[email protected]Distribute curated knowledge bases to associates. Revoke access when they leave. Audit every query.
Share proprietary methodology with clients on a project basis. Access expires when the engagement ends.
Distribute clinical knowledge bases to staff without exposing source documents or patient data.
Share indexed findings with collaborators while keeping raw datasets protected and access audited.
Full feature set — MCP server, OpenAI API, encrypted exports. No account. No telemetry. Runs entirely on your Mac.
Requires Ollama ·
ollama pull nomic-embed-text ·
Apple Silicon or Intel