Infino
★ star on github Contact an engineer

Supermodel helps frontier models retrieve the right data.

It queries your data more accurately, faster and with fewer tokens than a frontier model on its own. Switch your agent’s model to Supermodel and it runs on your data the same day.

Contact an engineer Get started for free
your agent or app question answer supermodel frontier model only what matters small models your bucket tickets.sf.parquet
infino eval

The same frontier model gets more answers right with Supermodel.

  • 17.7 pointsmore questions answered exactly right
  • 80%lower cost per question
  • 74 sa question, against 156 s

Databricks' OfficeQA Pro V2 benchmark

  1. 1 Supermodel + Opus 5.5 time per question: 74 s answered exactly right: 73.3%
  2. 2 Claude Opus 5.5 alone time per question: 156 s answered exactly right: 55.6%
  3. 3 GPT Astra time per question: 456 s answered exactly right: 49.6%
cat WHY.md

Why the answers are better and cheaper.

A frontier model on its own

  • Reads raw search results and whole tables, and pays for every token
  • Guesses what each column and code means
  • Joins tables on keys it hopes are right

The same model on Supermodel

  • Small models next to your data read the results first and pass on only what matters
  • Reads a card for every table, written when the data was loaded
  • Joins on keys the engine has already checked

where the cards and joins come from →

infino connect

Every team works from one copy of the data, under one set of permissions.

A PM or a CSM

POST /tickets/v1/messages

Asks in plain English and gets the answer back.

An engineer’s agent

search · find · sql · read · join_keys

Claude, Cursor or your own agent calls Infino’s tools over MCP.

A data team

hybrid_search('tickets', …)

Runs SQL with search inside it and reads Arrow or JSON.

An app on Elasticsearch

POST /tickets/_search

Sends the same Query DSL to a new endpoint.

cat request.http

Your agent keeps its code. Only the model changes.

Supermodel speaks the Anthropic Messages API and OpenAI’s chat-completions API. Point Claude Code, Cursor or an SDK at a base URL that names your database, send your Infino API key, and ask.

request.http
POST /tickets/v1/messages
x-api-key: $INFINO_API_KEY

{
  "model": "supermodel",
  "messages": [{
    "role": "user",
    "content": "Which plans filed the most disk-full tickets?"
  }]
}

Ask your own data a question.

Contact an engineer Get started for free

or run it yourself: pip install infino · Apache-2.0 · quickstart →