The Wren Journal

News, Product

TrueNAS × Wren AI: Private GenBI on Storage You Already Trust

Wren AI now runs next to the data you keep on TrueNAS. Ask in plain language, get governed SQL from a local model, and protect what the agent learns with OpenZFS snapshots and replication.

Wren AI Product Team

Updated: Nov 21, 2026
Published: Nov 21, 2026

TrueNAS × Wren AI: Private GenBI on Storage You Already Trust

At SC26 in Chicago this week, TrueNAS joined Phison and Wren AI as a hardware partner. The result is a simple idea with a big payoff for anyone who has to pass a security review: run GenBI where your data already lives. Wren AI sits inside your network, next to the data you keep on TrueNAS. Your team and your AI agents ask questions in plain language, a local model writes the SQL, and nothing leaves the building.

Two layers, one job each

TrueNAS is the storage and data-protection layer. It holds the datasets your databases and engines read, on OpenZFS storage you operate: end-to-end checksums, snapshots you can roll back to, replication to a second system, and file, block, and S3-compatible object access.

Wren AI is the GenBI layer on top. It reads your schema, builds the context model, and answers plain-language questions with SQL you can inspect. Row- and column-level security, validated SQL, and a full audit trail apply before any query runs, whether the question comes from a person in Slack or Teams or from an AI agent over MCP.

Both layers sit inside one network boundary. There is no cloud LLM in the loop, and the deployment can run fully air-gapped.

The model goes to the data

Most AI analytics tools work the other way around: send the prompt, the schema, and sometimes the rows to a hosted model and hope the contract covers it. On TrueNAS, a question flows like this:

  1. Storage. Databases, files, and object data stay on TrueNAS and are queried in place. No ETL, no migration.
  2. GenBI. Wren AI grounds the question in your schema, metric definitions, and permissions before any SQL runs.
  3. Inference. A validated open-weight model, served by an endpoint you run (such as Ollama or vLLM) or delivered on the Phison AI Data Platform, writes the SQL on your own GPUs.

Zero outbound calls, zero prompts shared, zero tokens metered.

What the agent learns is worth protecting

A GenBI agent does more than read. Over time it writes things your company comes to rely on, and on TrueNAS those get the same protection as the data itself:

Wren AI storesWhat it holdsOn TrueNAS
Context modelMetric definitions, joins, and access rulesSnapshot before every change; a bad edit rolls back to the last good version
Memory and skillsWhat the agent learns from daily use, and procedures your team savesReplicated to a second TrueNAS system so the knowledge outlives any one machine
Audit logsUser, timestamp, and SQL for every questionKept on checksummed OpenZFS storage for as long as your retention policy asks
Source dataThe databases and files Wren AI queriesServed over NFS, SMB, iSCSI, or S3-compatible object access

Snapshot, replication, and retention schedules are set in TrueNAS and scoped with your storage team during rollout.

What TrueNAS says

Organizations are looking for practical ways to bring AI closer to their data while maintaining control, security, and resilience. Our collaboration with Wren AI and Phison combines trusted business intelligence, accelerated AI infrastructure, and enterprise-grade data protection into a solution that helps customers deploy private AI with confidence.

Jamila Gunawardena, Director of Solutions and Alliances, TrueNAS

Start from the hardware you have

You do not need to buy a new stack to get started. Pick the path that matches what is already in your racks:

Get started

Keep the data where it is and start asking it questions. Request a demo and we will walk your data, storage, and security teams through Wren AI on TrueNAS, then scope a proof of concept on your own schema.

The Wren Journal

Get the next deep dive in your inbox

Practical GenBI guides, product updates, and customer lessons from the Wren AI team. A couple of emails a month, no noise.

Keep reading