Use case · Your model, your keys, your network
AI on your data, on your terms
Use AI to draft pipelines, explain failures and answer questions about your own data — with a model you choose, including one running fully offline, and with every tool scoped to the permissions of the person asking.
- Model
- Yours — local or your provider
- Offline
- Supported, first-class
- Copilot tools
- 26, permission-scoped
- Token metering
- None — we are not in the call
Who it is for
For teams who want the assistant without the new disclosure — whose schemas, metrics and sometimes rows must not be sent to a provider the vendor chose.
What this looks like from the inside
Every data tool now ships an assistant, and almost every one of them routes your prompts — which contain your schemas, your metrics and sometimes your rows — through a provider the vendor chose. Opting out usually means losing the feature.
Drawn, not described
The one thing the prose cannot do: show the moving parts and where the boundary is.
you
“why did extract_refunds retry?”
contains your schema · your run · maybe rows
Copilot
26 tools
scoped to your permissionsoption a · local model
ollama · this machine
no egressoption b · your provider
your account · your keys
your agreement, not oursNo token metering, no provider chosen for you. Without a frontier model it still works — just less fluently.
Step by step
In the order it actually happens.
- 01
Choose where the model runs
Point the Copilot at a local model such as Ollama on your own machine, or at the hosted provider you already have a data agreement with, under your own account and keys. Revoluzen does not resell tokens or sit in the middle of the call.
- 02
Draft a pipeline from intent
Describe what you want — pull orders from Postgres, conform with refunds from Stripe, load to gold. The Copilot drafts the DAG on the canvas; you review it like any other change.
- 03
Ask why a step failed
The Copilot reads the failed step's inputs, outputs and Steward findings and explains the failure in terms of your pipeline, not in generic advice.
- 04
Ask about your instance, within your permissions
All 26 tools are read-scoped to your pipelines, runs and catalog, and they respect the permissions of the person asking. The assistant cannot see more than you could open yourself.
- 05
Let assistants query the governed model
Revoluzen Studio's governed MCP server exposes defined metrics and the semantic model — not raw tables — so an assistant answers business questions with governed answers.
What actually changes
| Dimension | Today | With Revoluzen |
|---|---|---|
| Where prompts go | To a provider the vendor chose, under the vendor's agreement. | To a model on your machine, or to your provider under your keys. Never through Revoluzen. |
| What opting out costs | The feature. | Fluency. With a small local model the Copilot drafts less well; it still works. |
| What the assistant can see | Often more than the person using it. | Exactly what the person using it is permitted to see. |
Which product does what
Each product is adoptable on its own; together they share one catalog, one permission model and one run history.
Revoluzen Flow
The Copilot drafts pipelines from intent, explains a failed step and answers questions about your instance, through 26 tools scoped to the asking user's permissions.
Product pageRevoluzen Studio
A governed MCP server lets assistants query defined metrics and the semantic model instead of raw tables.
Product pageWhere it runs
Point the Copilot at a local model such as Ollama and no prompt leaves the machine — a supported first-class mode, not a degraded fallback. Or use the hosted provider you already have an agreement with, under your own keys. Revoluzen does not resell tokens or sit in the middle of the call.
What you end up with
- 01AI features that work with no frontier model at all, just less fluently
- 02Prompts that stay on your network, or go only where you have an agreement
- 03No vendor token metering, and no provider chosen for you
- 04An assistant that cannot see more than the person using it
The honest limits
You would find these out eventually. Finding them out now is cheaper for both of us.
Quality of drafting depends on the model you choose; a small local model drafts less well than a frontier one.
Revoluzen is not an AI company and does not try to be. AI is a feature inside the products.
The governed MCP server is part of Revoluzen Studio, which is in early access.
Asked about this scenario
Which AI models can Revoluzen use?
Any you choose. Revoluzen's Copilot can run against a locally hosted model such as Ollama, so no prompt leaves your machine, or against a hosted provider using your own account and keys. Revoluzen does not route prompts through its own infrastructure.
Can Revoluzen's AI features run with no internet connection?
Yes. With a local model, Revoluzen's Copilot runs entirely offline, and the products function fully without any frontier model.
Can Revoluzen's AI assistant access data the user cannot?
No. All 26 Copilot tools are scoped to the asking user's permissions, so the assistant cannot read a pipeline, run or dataset that the person using it could not open directly.
Adjacent scenarios
Dashboards without the export
On-premises analytics with governed answers
Dashboards, paginated reports and a governed query layer that run inside your network against your own warehouse — where every number on a screen can say which metric and which query produced it.
Keep data in-country, in-network
Data residency and sovereignty
Run pipelines, a lakehouse and analytics entirely inside your own network — with no hosted control plane, no telemetry and no outbound dependency — so residency obligations are met by architecture rather than by contract.
Run it yourself, then decide.
Revoluzen Flow is Apache-2.0 and installs in one command. If it fits, we can talk about the parts that are not open source.