Foundational AI Research Lab for Efficient Language Modeling

Frontier models should not readEVERYTHING.

An upstream relevance filter for agent stacks. Remove irrelevant context before it reaches the language model. Cut token cost. Keep answer quality and your existing workflow.

Get API Key

context filter

Query

tariffs and trade policy

Chunks

keepTariffs on imported steel may lift COGS by $50M.
keepTrade-policy shifts hit China-sourced lines.
dropOur company picnic was wonderful this summer.
dropSame-store sales grew 4% year over year.

2 of 4 kept

Kolmo

Extreme compression for relevance.

Named for Kolmogorov complexity.

Kolmogorov complexity asks for the shortest program that produces a given output: how much structure you can pack into how little. Kolmo is our Tiny Language Model family built in that spirit. Sub-100M parameters. Zero-shot relevance filtering. Upstream of the frontier model so multi-million-token corpora shrink to the evidence that matters, without rewriting the agent workflow.

A filter only earns its place if it holds up in production, not just on a leaderboard. Kolmo is designed so relevance stays sharp while cost and latency drop, decisions stay replicable for audit and compliance, and less unnecessary context reaches the prompt.

Massive reasoning context. Minimal model footprint.

Effortless integration

Drop ZQ in front of your model via API or MCP. Full reference lives in the docs.

REST API

Filter chunks in one call

Send text chunks and a topic or query. Get back only what is relevant, with scores and token accounting.

predict

curl -X POST "$ZQ_BASE_URL/v1/vslm/predict" \
  -H "x-api-key: $ZQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent": "general_context_agent",
    "query": "tariffs and trade policy",
    "sentences": ["...", "..."]
  }'

API docs

MCP

Use it inside your agent host

Add the ZettaQuant MCP server to Cursor, Claude, or Zed. Use the full path from which uvx, set ZQ_API_KEY, restart the host.

mcp.json

{
  "mcpServers": {
    "zettaquant-vslm": {
      "command": "/opt/homebrew/bin/uvx",
      "args": ["zettaquant-vslm-mcp"],
      "env": { "ZQ_API_KEY": "YOUR_KEY" }
    }
  }
}

MCP setup

Try it in minutes.

Grab an API key and paste the onboarding prompt into your agent.

Get API Key