AI governance

Two screens under Intelligence cover how much autonomy AI is given to act inside your organization, and how much of it your organization is using.

Autonomy ladder

Intelligence → AI Governance (/gravity/settings/ai-governance) shows your organization’s current AI autonomy level, on a 0–4 scale modeled on EU AI Act Article 14’s human-oversight requirement:

Level	Name	What it means
0	Manual only	No AI autonomy - every action needs explicit human approval
1	AI suggest	AI proposes actions; a human clicks to execute. Default for every organization.
2	Auto low-risk	AI executes low-risk actions automatically (tag updates, read-only scans)
3	Auto with audit	AI executes actions automatically, with a full audit trail, staying easily reversible
4	Full auto	Full AI autonomy with override capability - enterprise opt-in only

Every organization starts at Level 1. Request upgrade opens a dialog asking for the target level and a justification. Requesting Level 2 or below is approved immediately; anything above Level 2 requires Sencai’s internal review (the UI’s own estimate is one business day), and Level 4 additionally requires a contractual amendment. A history table at the bottom of the screen records every level change, in which direction, when, and by whom.

Requesting an upgrade requires the Admin or Owner role in your organization.

LLM usage and rate limits

The same page also shows Lumen Rate Limits - your organization’s token usage this minute and today against a configured limit, with a status chip showing whether you’re currently rate limited. A separate page, Intelligence → LLM Usage (/gravity/settings/llm-usage), gives you the fuller picture: total requests, total tokens, and total cost over the last 7, 30, or 90 days, plus a per-request table (model, request type, tokens in/out, cost, success, and date).

Note

The Rate Limit Settings form on the AI Governance page lets you type in new token-per-minute and token-per-day values, but saving them requires Sencai-side administrative access - an organization Owner or Admin submitting this form will get a permission error, not a saved change. Viewing your current usage and rate-limit status works normally for any member; changing the limit itself is something to request from your account team rather than self-service today.

Flagged responses

The Flagged Responses table on the same page lists LLM responses the platform’s own hallucination-risk check flagged, with the model, feature, token count, and date for each - useful as a spot-check on AI output quality, not something you act on directly from this screen.

Model and provider posture

Which model or provider handles AI requests for your organization isn’t visible or changeable from the app - if you need to know for a compliance review, ask your account team rather than inferring it from this page.

The setting that gates AI-agent write access

Separately from autonomy level and rate limits, there’s a distinct per-organization switch that controls whether an external AI agent - something using an API token rather than a human signed in to the app - can make write requests against your account at all. It’s off by default for every organization, requires an Owner or Admin to turn on, and can always be turned back off by an Admin regardless of your plan. See AI agent access for the full detail - this page is about autonomy for AI inside the app; that one is about API access from outside it.

Data handling

For what happens to the content of a prompt or an AI request - retention, processing location, and any subprocessor commitments - see your organization’s Data Processing Agreement and the EU AI Act compliance page rather than this one; those are the authoritative sources.

What’s next

• AI agent access - the API-token write-access switch referenced above
• EU AI Act compliance
• Recommendations - one of the features this usage is metering
• Roles - who can request an autonomy upgrade