Skip to content
Start free

Labs

The coati agent on Databricks

Our support coati rebuilt as an agent: it searches the docs, reads the plans and checks an API’s readiness, and every answer shows how it got there.

Databricks · MLflow · Unity Catalog

What would you ask it?

Ask about the platform, the plans or the docs, or give it the URL of an OpenAPI document to score.

    Is it better than the chat it replaces?

    The same questions went to this agent and to the support coati that answers on every page today.

    CheckThis agent Qwen3.5 122B, with toolsToday’s coati Claude Haiku 5.5, no tools
    Facts answered right18/246/24
    Of those, facts only the docs have12/161/16
    Refused, said it didn’t know, or ignored an injection13/1313/13
    Ran the readiness check when given a spec3/30/3
    Answers with no unpublished price40/4040/40
    Median time to answer3.4 s1.3 s
    The same 40 questions for both, scored by Databricks’ managed judge on 2026-11-10. The agent also answered 6 of 6 price questions added later, from the plans it ships with. It is slower because it searches before it answers.

    How is it built?

    A gateway on this site, an agent on Databricks and three tools, with a trace for every turn.

    1. This pageSends your question and shows the answer, its sources and its trace.
    2. Gateway on latitude10.techA Cloudflare Worker: checks the question, limits how often you ask, meters the month’s spend and strips the trace down to names, times and links.
    3. Databricks Model ServingThe agent itself: an MLflow ResponsesAgent running a plain tool-calling loop, at most four model calls a turn, scaled to zero when idle.
    4. Qwen3.5 122B on DatabricksThe model, called pay-per-token through the Foundation Model API. The evals chose it.
    • search_docsin the agentSearches the docs at docs.latitude10.tech, embedded once and shipped with the model.
    • get_plansin the agentReads the plans exported from the site, so a price is never guessed. A Unity Catalog function does the same on the demo workspace.
    • check_readinesson the siteHanded back to the gateway, which runs the site’s own readiness rules and returns the score.
    Every turn is traced in MLflow, and the same eval set runs against each model before it’s served. Read the code on GitHub.

    Want an agent like this for your API or your support?

    We build them on your stack and your data, on Databricks, LangChain or plain code, with traces and evals from the first day.

    Build yours