Auth

Roadmap

Token issuance scoped to specific users, with audit-log accountability for every agent action.

Auth models

Why on-behalf-of-user matters

The naive model is to issue a tenant-wide API key to the agent and let it do anything. That fails procurement review at any serious mid-market SaaS. The correct model is: each deal-desk user can authorize agents individually, and the agent’s scope is the union of the user’s scope and the authorization grant. A rep’s agent can draft a quote and propose a 10% discount, but it cannot auto-approve a 30% discount that requires VP sign-off. That still routes to the human.

Audit accountability

Every action taken with an agent-on-behalf-of-user token records the tuple (agent_id, on_behalf_of_user_id, action, deterministic_inputs, computed_outputs) in the audit log. A SOX-style review can reconstruct who (human or agent) did what, with which inputs, producing which outputs. The deterministic boundary means the “outputs” replay reproduces the same numbers months later.

Scope model (planned)

What’s next

Auth shape is being scoped with the same design partners working on webhooks and the MCP server. If you’re building a custom Slack agent and need on-behalf-of-user auth before GA, talk to us.