OPEN SOURCE · SELF-HOSTED · APACHE-2.0

Open source, managed Trino clusters,
in your own AWS account.

Provision EC2-backed Trino clusters, wire up catalogs, and run SQL — all from one box. A Trino-aware autoscaler and auto-suspend tear down idle capacity automatically, so you pay for queries — not idle clusters.

Try the live demo Deploy on AWS View on GitHub

Sign in to the demo as demo / trinohub-demo — read-only, on sample data. Browse clusters, catalogs, the SQL editor, notebooks, Ask Trino and MCP, and run real SQL against tpch and tpcds. The cluster is suspended, so your first query spends a minute or two starting it.

Runs in your VPC No static AWS keys, ever FastAPI + boto3 + SQLite
trinohub.internal / clusters
TrinoHub clusters view
3 tiers
Cost · Balanced · Power presets
1 stack
One CloudFormation deploy
0 keys
Instance-profile auth only
S3 + Glue
Iceberg catalogs, your data
Trino-aware autoscaling

Pay for queries,
not idle clusters.

A control loop — not a raw CPU policy — scales workers on queued queries and CPU, with cooldowns and min/max bounds. Idle clusters auto-suspend, and a query against a suspended cluster resumes it first.

  • Scale on real query demand, with cooldowns to avoid thrash
  • Auto-suspend tears down idle capacity to stop the meter
  • Optional NVMe caching keeps hot S3 data on local disk
Cluster detail — metrics, cost, and live utilization
trinohub.internal / ask-trino
Ask Trino — plain-English questions answered with SQL, tables, and charts
Ask Trino · AI assistant

Ask in plain English.
Review the SQL. Run it.

Pick a cluster, catalog, and schema, then ask a question in everyday language — or start from a suggested prompt. Ask Trino drafts the SQL with a short explanation, and one click sends it to the SQL editor to run against the same cluster.

  • Only emits SQL as text — nothing runs until you say so
  • Read-only by design: a single SELECT — writes and multi-statements are rejected
  • Bring your own LLM provider — API keys stay in server env vars, never logged, never sent to the browser
MCP server · Built in

Point Claude at your
own Trino clusters.

TrinoHub serves a Model Context Protocol endpoint at /mcp over streamable HTTP, so Claude Code and other MCP hosts can list clusters, browse metadata, and run queries — against the same clusters, catalogs, and permissions your analysts already use.

  • Eight tools, in four groups
    Discoverylist_clusters browse_metadata Queryrun_query get_query_result Templateslist_query_templates run_query_template Data productssearch_data_products get_data_product
  • Not just SELECT — clients run SHOW, DESCRIBE, and EXPLAIN too, which is what makes schema discovery and plan inspection work. EXPLAIN ANALYZE stays refused — it executes what it explains — and EXPLAIN re-validates whatever follows, so a write cannot ride in behind the keyword. Ask Trino remains SELECT-only.
  • Results capped at 100 KB, separately from the browser's 1,000-row / 10 MB display cap — a tool result is read into a model's context, not rendered in a table. Trimmed results still carry a truncated: true flag.
  • Typed results: tools declare an outputSchema and return structured content — no text scraping
  • Read-only and non-destructive hints keep hosts from prompting on every call
  • Bearer-token auth with OAuth discovery: RFC 9728 metadata at /.well-known/oauth-protected-resource, and an unauthenticated /mcp call answers with a WWW-Authenticate challenge pointing at it. Protocol revisions are negotiated per client — not hardcoded.

The in-app MCP page hands you the endpoint, transport, and auth header as copy-able fields — plus ready-made setup for Claude Code, JSON-config clients, and curl.

bash CLAUDE CODE 
claude mcp add --transport http trinohub \
  https://trinohub.internal/mcp \
  --header "Authorization: Bearer $TRINOHUB_TOKEN"
Protocol revisions 2025-03-26 2025-06-18 2025-11-25
Everything from one box

The control plane for your Trino fleet

A single EC2 instance runs the UI, API, and AWS orchestration. From there, operators launch clusters and analysts run SQL — no Terraform, no Kubernetes.

Cluster lifecycle

Create, start, suspend, disable, and delete EC2-backed clusters. Delete terminates every tracked AWS resource.

Trino-aware autoscaling

Scales workers on queued queries and CPU with cooldowns and bounds. Auto-suspend stops idle spend.

Catalogs: S3 + Glue

Built-in system, tpch, tpcds, plus S3 + AWS Glue Iceberg. Worker IAM grants access — no S3 keys stored.

SQL editor

Schema browser, syntax highlighting, live status polling, tabular results, and streaming CSV export.

Notebooks

Ordered SQL cells with inline tables or charts, per-cell overrides, Run all, and autosave to your account.

Accelerated caching

Optional warm cache on local NVMe for S3-backed Hive, Iceberg, and Delta — repeated scans read from disk.

Data products

Documented bundles of tables and views — owner, tags, a description per asset — plus named SQL with typed {{placeholders}} that callers fill with values, never SQL. Agents find them by keyword instead of guessing from raw catalog metadata, and publishing one describes data without widening who can read it.

Query history

Every run with status, cluster, user, elapsed, and rows. Open a row for the full SQL, the catalog and schema context, the Trino query ID, and a per-stage execution breakdown.

Scheduled jobs

Run one SQL statement on a cron or every-N-minutes schedule. A failed run retries once, and a job can run as a service account instead of you — with its own run history.

Connectors

26 connectors out of the box

Every connector ships with TrinoHub — the picker, catalog forms, and credential handling all derive from a single registry. Add a catalog from the UI and it's queryable in seconds.

Object storage4

S3 + Glue catalogs authenticate with IAM roles — no stored secrets.

S3 + Glue (Iceberg)Delta LakeHiveHudiquery-only

Document & search3

Query document stores and search indexes with plain SQL.

MongoDBElasticsearchOpenSearch

Databases & warehouses14

JDBC connectors configured from the UI — point at a host, add credentials, query.

PostgreSQLMySQLAmazon RedshiftSQL ServerMariaDBSingleStoreClickHouseSnowflakeOracledriver uploadTeradatadriver uploadTrino (cross-cluster)plugin uploadApache DruidApache CassandraPrometheus

Google Cloud2

Authenticate with a GCP service-account JSON key.

BigQueryGoogle Sheets

Test & sample data3

Zero-config catalogs for trying queries before wiring up real data.

MemoryBlack HoleFaker
Security by default

Your data never leaves your account.

TrinoHub installs into your own VPC. Query data flows between your worker nodes and your S3/Glue over your own network and IAM — nothing routes through a TrinoHub-hosted service.

No static AWS credentials, everInstance-profile auth for the control plane, a passed node role for clusters. The UI never displays keys.
Deploys into your own VPCControl plane and every cluster live in your account. Allowed-UI-CIDR enforcement at the app layer.
Signed per-cluster bootstrap tokensEach node fetches only its own config. Hashed passwords and httpOnly session cookies throughout.
How it works

From zero to SQL in three steps

1

Deploy the control plane

One CloudFormation stack creates the IAM roles (instance-profile auth — no static keys), a security group, and an EC2 instance that installs and starts the app.

bash 
aws cloudformation deploy \
  --stack-name trinohub \
  --template-file deploy/aws/cloudformation.yaml \
  --capabilities CAPABILITY_NAMED_IAM

Prefer Terraform, or CloudFormation restricted in your account? The control plane is a plain FastAPI app behind nginx — install it on any Ubuntu 24.04 host you manage →

2

Launch a cluster

Pick a preset tier — Cost, Balanced, or Power — toggle autoscaling and auto-suspend, wire up catalogs, and hit create. Each cluster maps to a tagged set of AWS resources.

Create cluster — preset tiers, autoscaling, and review panel
3

Run SQL

Analysts pick a cluster and catalog, write SQL with a live schema browser, watch query status, and download CSV — or organize it all into notebooks.

SQL editor — schema browser, results, and recent queries
TrinoHub

Open source. Self-hosted. Free.

Clone the repo, deploy the stack into your AWS account, and run the setup wizard. That's it.