Getting started

Configure an agent and enforce policies, end to end. Configuration is admin-only; once set, enforcement happens automatically on every action.

1. 1

   Create a Mandate — the agent's scope of authority

   Mandates → New mandate (or define one inline when registering the agent). A mandate sets the allowed action types & resources, a hard per-action value cap, the value above which approval is required, a rate limit, and optional permitted hours. Mandates are reusable across agents. See Agents & mandates and Resources & naming.
2. 2

   Register the Agent

   Agents → Register agent. Set the name, environment and tags, pick an owner (a team member who becomes the agent's default approver), and attach the mandate. Copy the API key — it is shown once; store it as PROVENANT_KEY in the agent's environment (rotate any time from the agent's page).
3. 3

   Set a Budget — the spend firewall

   Budgets → New budget. Scope to the org, an agent, or a tag; set the limit, the window (rolling_24h / daily / weekly / monthly / total), and the behavior on breach: block or require approval. Spend is reserved on authorization and reconciled to the actual amount on completion.
4. 4

   Create Policies — org-wide rules

   Policies → New policy. Policies layer on top of every mandate, matching on action type, resource, tags, environment, value range, risk band, time window or metadata, and contributing an effect (allow / require_approval / deny). Stage them draft → shadow → active to observe before they enforce. See Policies. Your org already ships with starter guardrails — including a "Block sanctioned counterparties" deny — so you can see Provenant say no in your first session: authorize a payment.send to vendor:sanctioned-demo and it comes back denied with the reason. Edit or remove any of them on the Policies page.
5. 5

   Connect your agent and verify

   Wire the SDK into your agent (Connect the agent), then watch decisions land live in Activity and verify the chain in the Audit ledger. Preview any hypothetical action first with the simulator on an agent's page.