Skip to main content

The AI/BI layer Trino has been missing

· 4 min read
Actyze Team
Actyze Team
AI-Powered Database Analytics

Trino is one of the most important pieces of modern data infrastructure. It lets you treat PostgreSQL, Snowflake, MongoDB, BigQuery, Databricks, Iceberg, and a dozen other systems as a single federated query surface. If you've adopted it, you've solved one of the hardest problems in analytics: putting the data within reach of one query plan.

What Trino has never had is a great query and visualization layer purpose-built for it.

The shape of the gap​

Most BI tools were built before federation went mainstream. They assume one database, one schema, one connection — and then bolt on multi-source support as an afterthought. Trino exposes a single SQL surface across catalogs, but the BI tool fights it: dashboards become single-source, semantic layers don't traverse catalog boundaries, and joins across systems require manual modeling.

The newer wave of AI analytics tools has the opposite problem. They're AI-native — natural language to SQL, conversational refinement, even voice — but they're built for a single warehouse. Snowflake Cortex stays inside Snowflake. Databricks Genie stays inside Unity Catalog. Neither speaks to your federation reality.

If you've gone to the trouble of running Trino, you don't want either tradeoff. You want both: AI-native conversational analytics, and the federation you already built.

What Actyze is​

Actyze is an open-source, self-hosted AI analytics platform with Trino at its core. The shape of it:

  • Natural-language → SQL across federated catalogs. Ask in 50+ languages. The query plan goes through Trino, so the answer can join data sitting in Postgres against data sitting in Snowflake against data sitting in MongoDB — in a single question.
  • A living semantic relationship graph. Joins are inferred from foreign-key conventions, mined from successful query history with sqlglot, and verified by admins. The LLM doesn't hallucinate JOIN conditions — it reads the graph.
  • No-code ML predictions. Forecast, classify, estimate, and anomaly detection run on Trino-queryable data. AutoGluon, XGBoost, and LightGBM are picked automatically based on the outcome shape.
  • Scheduled KPIs as a gold layer. Define a SQL query, set a schedule, get a materialized table. No ETL pipeline to maintain.
  • Voice analytics. Web Speech API and Whisper for hands-free or accessibility-first querying.
  • Bring your own LLM. 100+ providers via LiteLLM — Anthropic, OpenAI, Gemini, Llama, Ollama. Your key, your model, your network.

It's AGPL v3, no feature gates, no per-credit pricing, no proprietary cloud. Deploy with Docker Compose, Helm, or Kubernetes — including air-gapped — in minutes.

Why this matters now​

Three things are converging:

  1. Federation has won the architecture argument. Most large data teams have given up on the "single warehouse" dream. Trino, Starburst Galaxy, AWS Athena Federated Queries — the federation layer is becoming foundational, not exotic.
  2. AI-native analytics is the new default for self-serve. Asking the data in plain English is no longer a science project. The question is which AI surface to put on top of your stack.
  3. The proprietary AI/BI vendors are pulling in the opposite direction from federation. Cortex deepens lock-in to Snowflake. Genie deepens lock-in to Databricks. The more they invest, the worse the federation story gets.

If federation is foundational and AI is the new default, then the AI layer needs to live on top of federation — not inside one warehouse. Actyze is the open answer to that.

Try it​

Spin up Actyze against a Trino cluster you already have, or use the bundled setup that runs Trino as part of the deployment.

git clone https://github.com/actyze/dashboard.git
cd dashboard/docker
cp env.example .env
# add your LLM API key
./start.sh

Frontend at http://localhost:3000, API at http://localhost:8000. Point it at your Trino cluster's catalogs, ask a question that crosses two of them, and watch the AI write the SQL.

Where to find us​

If you're running Trino at production scale and want to talk about what an AI-native federation layer should look like, reach out. We're always looking to learn from teams who've been deeper in this than we have.