Tooli is the planning assistant built into Toolio. It answers questions about your plans and can propose edits, and it only ever sees the data and takes the actions that the signed-in planner already has in Toolio. This article explains how Tooli works with your data, where that data goes, how long it is kept, and what your team controls.
Use Cases
Answering a security or privacy review before your team turns Tooli on
Explaining to IT or legal which AI provider Tooli uses and whether your data trains models
Checking how long Tooli keeps conversations, files and traces
Deciding which Tooli features to enable for your team
At a Glance
Tooli runs on Anthropic's Claude through Anthropic's commercial API. Those terms prohibit training on your data, and Toolio does not use it to train general-purpose or generative AI models either.
Tooli works with plan, assortment, item and allocation data. Toolio holds no consumer data, no payment card data and no sensitive personal data. The only personal data is each planner's business email address, used for sign-in.
Every request runs as the signed-in planner, through the same permissions, role-based access control and datascope as the Toolio app. The model has no direct access to your database.
The Toolio platform is hosted on Microsoft Azure in the United States (US-East). The model and the agent session run at Anthropic.
Tooli is turned on or off for each environment at your request.
How Tooli Works
The model never touches your database. It gets data by calling Toolio's MCP server as the planner, so the same permission checks and audit log apply as in the app.
A planner asks Tooli a question in Toolio.
Tooli starts an agent session with Claude at Anthropic and sends it the question.
Claude requests the data it needs by calling Toolio's MCP server with the planner's session credential. Toolio checks permissions and datascope and returns only what that planner can see.
The MCP server reads from and writes to your own tenant database. Edits in chat wait for the planner's approval, and every change is logged.
Prompts, answers, tool inputs and outputs, and the planner's email are traced in Braintrust for quality and incident review.
What Data Goes Where
Plan data leaves Toolio's Azure environment only to answer a planner's question, and nothing is used to train models.
Data | Where it is stored | How long | Who can see it |
Plan data used to answer a question | Sent to Anthropic for the model call | Under Anthropic's commercial API terms, never used for training | The planner who asked |
Conversation list and details | Toolio, Azure US-East | 30 days after the last message | The planner, plus anyone they share a read-only link with |
Conversation transcript | Anthropic | 30 days after the last message | The planner, plus anyone they share a read-only link with |
Team memory (off by default) | Anthropic, one memory store per organization | Until edited or deleted; prior versions kept 30 days | Everyone in your organization, once enabled |
Files a planner attaches | Anthropic | Deleted with the conversation, 30 days after the last message | The planner |
Traces of prompts, answers, tool inputs and outputs, and the planner's email | Braintrust | About 5 weeks | Toolio engineering, for quality and incident review |
Changes made to the plan | Toolio, Azure US-East | Same as any plan edit | Change History and audit log: who, when, old and new values |
If you connect your own AI client to Toolio through the MCP server instead of using Tooli, Toolio sends nothing to an AI provider. Your client's own terms apply.
Access and Controls
Toolio enforces permissions outside the model. A planner who writes any prompt they like still sees only what they could see in the app, and can take only the actions they already have rights to.
In chat, Tooli calls the MCP server with the planner's own session. Scheduled and Slack agents use a key limited to that agent's role. Every call passes the same role-based access control and datascope checks as the web app.
In chat, every edit Tooli proposes appears as an approval card, and nothing is saved until the planner accepts it. Admins set each agent's tools to ask, always allow or disabled. Historical actuals such as sales, inventory and receipts cannot be written by Tooli.
Tooli is turned on or off per environment. Team Memory is off by default and needs an admin to turn it on. MCP access can be set to Enabled, Disabled, or Restricted to specific roles.
Monthly usage caps apply per user and per organization, and the MCP server rate-limits each user.
Tooli and the MCP server run inside the production environment covered by Toolio's SOC 2 Type II audit, with annual third-party penetration testing.
Recommended Rollout
If your security team wants to review Tooli before a full rollout, we recommend starting small:
Your security and privacy teams review this article and Toolio's SOC 2 Type II report, available in the Toolio Trust Center.
Toolio enables Tooli in a pre-production environment for a small group of named planners.
The pilot runs in chat only, with tools set to ask, Team Memory off and no scheduled or Slack agents.
Toolio shares usage and a sample of the audit log at the end of the pilot, and you decide on production.
FAQs
Does Toolio or Anthropic train AI models on my data?
No. Anthropic's commercial API terms prohibit training on your data, and Toolio does not use it to train general-purpose or generative AI models. Toolio's forecasting, curve and clustering models learn from your data only to produce your own plans, and are never shared with other customers. See Anthropic's answer for its terms.
Can Tooli see data a planner does not have access to?
No. Tooli runs as the signed-in planner, and Toolio checks that planner's permissions and datascope on every request before returning any data.
Can we turn Tooli off?
Yes. Tooli is enabled per environment at your request, and you can ask your Toolio team to turn it off at any time. Team Memory and MCP access have their own controls.
Where can I find Toolio's sub-processors and SOC 2 report?
Both are in the Toolio Trust Center. The SOC 2 Type II report is available on request there.