Self-hosted RAG · Runs on your infra

RAG without
the data leak.

Upload documents, ask questions in plain English, get cited answers — all on your own server. OpenAI never sees a byte.

< 2 min to first answer
100% data stays local
0 third-party AI calls
Vector + BM25 hybrid search
Runs entirely on your server
PDF, DOCX, TXT supported
powered by
Docker
Redis
Postgres
Ollama
BullMQ
open source
built for

Teams that can't send
client data to OpenAI.

Legal teams

Contract review, clause extraction, matter research — without sending client files to OpenAI.

Research teams

Query across hundreds of papers. Find the passage you half-remember without re-reading everything.

Support teams

Ground your support bot in your actual product docs. No hallucinated feature names.

Developers

Embed a fully-featured RAG API into your product. No ML infra to manage.

// quick start

One endpoint.
Infinite answers.

The /ask endpoint accepts a question and optional file IDs. It retrieves the most relevant chunks via hybrid search, builds a grounded prompt, and streams back an answer with source citations — entirely on your own server.

streaming SSE source citations local LLM metadata filters
ask.ts typescript
// Ask a question across your documents
const stream = await fetch("/api/v2/files/ask", {  
  method:  "POST",
  headers: { "X-API-Key": process.env.KYRO_KEY },
  body: JSON.stringify({
    question: "What are the payment terms?",
    fileIds:  ["uuid-1", "uuid-2"],
    topK:     8,
  }),
});

// Response streams SSE events:
// { type: "sources",  sources: [{ fileId, score, content }] }
// { type: "chunk",    text: "According to [1], payment..." }
// { type: "done" }
platform

Built different by design.

RAG

Ask Your Documents

Ask a question, get a grounded answer with inline citations. The LLM only sees what's in your files — no hallucination, no guessing.

SEARCH

Hybrid Semantic Search

BM25 keyword search fused with pgvector similarity via RRF. Finds the right chunk even when the query doesn't share exact words with the source.

INFRA

Fully Self-Hosted

One docker compose up. Postgres, Redis, Ollama, and Kyro on your own server. Your documents never leave your network.

PLATFORM

API-First & Multi-tenant

Scoped API keys, role-based access, webhooks, collections. Embed it in your own product or wire it up to an internal tool in an afternoon.

getting started

Up and running in minutes.

Deploy in one command

Postgres, Redis, Ollama, and Kyro running locally. No cloud accounts, no API keys, no dependencies outside your machine.

docker compose up -d

Upload your documents

PDFs, DOCX, plain text. Text extraction and vector embedding run automatically in the background via BullMQ workers.

POST /api/v2/files

Ask questions via API

Send a question. Get a streamed, cited answer grounded in your documents. Wire it to a chatbot, a search UI, or a CLI.

POST /api/v2/files/ask
Free to self-host · MIT licensed · always open source

Your knowledge base.
Your rules.

No credit card. No vendor lock-in. No data egress. Your documents stay on your server, forever.

docker compose up -d