Why teams choose
AISO.
Most enterprises want AI on their most valuable data and can’t allow it. AISO governs who reaches what, lets you keep any model, and records every decision.
You choose the models, and they stay chosen.
Any provider on your own key, open-weight models on your own machines, or both, with a different model per agent, per project or per level of sensitivity. A pinned model can’t be changed by a vendor update or a per-request override, so an agent you tested keeps behaving as tested.
Open-weight models cut both the risk and the bill.
Qwen, Llama and Gemma run on MLX Serve, Ollama or LM Studio on Mac, Windows or Linux machines you own, and cost the hardware rather than a price per token. One server setting forbids cloud models outright. We run Claude Code through AISO on Qwen 9B on a Mac.
Sensitive data reaches only the models cleared for it.
File permissions are only as good as the last cleanup. AISO adds a gate that doesn’t depend on them. Documents, data sources and folders of code carry a level, and every request is checked against the person’s clearance and the model’s ceiling. When someone needs something they can’t see, Chat offers to ask its owner.
Every figure can be traced and defended months later.
Each number in a reply is matched against the query results that produced it, and any it can’t trace is named. A kept answer is signed with those results, verifies offline and can be re-run, on Snowflake and Databricks against the data as it stood that day.
Data stays in its source, read as the person asking.
AISO asks Snowflake, Databricks, Power BI, dbt, Looker or Tableau at the moment of the question, through each vendor’s own MCP server, signed in as the person. The row and column rules your data team set still apply, and nothing is copied into a second index.
Developers keep their coding tools.
Claude Code, Codex, Copilot and open-source extensions reach models only through AISO, as the developer, under a project’s models and budget. Code from a restricted folder is refused on cloud models whichever tool sent it, and CI reports which lines of a pull request a model wrote.
One rulebook, and a log that proves itself.
The same roles, approvals, classification and audit log cover the person in Chat, the builder whose agent answers and the developer whose tool calls a model. Every row is hashed into a chain, so an edit after the fact shows.
A proven setup spreads as one pack.
A pack carries agents, knowledge stores, data connections, workflows, schedules, roles and classification policy. It installs from any git repository and is undone if any step fails. You choose which publishers to trust, and code from an unknown one runs cut off from the network and your data.
Classification included
The same work produces the written policy, labels and evidence an ISO 27001 or SOC 2 review asks for.
A policy, drafted with you
An onboarding agent interviews an admin and drafts a policy shaped to ISO 27001 controls A.5.12 and A.5.13, or reads the criteria out of one you have. Nothing changes until an admin adopts it.
Classified on your own machines
The classifier reads each document as it is indexed and must run on a local model. Until it decides, a document is hidden from everyone below the top clearance.
A person approves anything lowered
The classifier can raise a level on its own, but lowering one waits for a person.
Watch, then enforce
Monitor mode records every read the policy would have refused while everyone keeps working, so you switch to enforce once that list looks right.
Handling rules, checked
Each level states who may read it, whether it may reach a cloud model and how long it is kept. The compliance check compares that with the live setup and names each fix. Every run it records goes into the audit log and the evidence bundle.
Already on Purview
Purview label names map to AISO levels, either as the final word or as a floor the classifier can raise.
See it on your own machines
We’ll run AISO against your identity provider and a slice of your data, then scope a deployment. Two weeks to production, or your money back.
Book a Demo