Documentation
Agentcess documentation: passports, capabilities, policies, zones, the gate, budgets, approvals, evidence and delegation.
Everything in Agentcess revolves around one idea: an agent is a governed actor. The concepts below are the whole model; once they make sense, every page in the app does too.
Core concepts
- Passport — the agent's portable identity: purpose, owner, status and a permanent identity history. Every change is a numbered record.
- Capabilities — named abilities bound to an agent, each with a risk tier: read, write, destructive or consequential.
- Access grants — exactly which data assets an agent may reach, and at which classification.
- Policies and policy packs — versioned rules, bundled into reusable packs, that gate actions in observe, warn, approval or block mode.
- Zones — where data must live and what may leave it. The gate refuses or escalates actions that would breach a zone's egress terms.
Running agents
- Runs — give an agent a task and a dollar budget. It executes on the server and keeps going if you close your phone.
- The gate — every action passes identity, capability, policy, zone and budget checks in a fixed order, and the decision is recorded.
- Approvals — actions needing a person land in Approvals, bound to the exact action and resource, verified by passkey.
- Delegation — an agent may hand part of its authority to another agent, but never more than it holds.
Evidence and oversight
- Evidence ledger — every interaction is recorded and attributable.
- Record book — history is folded into a hash-chained book you can export and verify independently.
- Behaviour monitor — refusal rates, spend and anomalies per agent, with automatic hold on repeated refusals.
- Security scan — a workspace-wide check for over-powered agents, open-ended clearances and risky tools.