Lifecycle Stages
draft → active ↔ archivedDraft: Being configured. Not deployable. Prompt, skills, tools, and governance settings under construction.
Active: Available for deployment and execution. The primary operational state. Editable with audit tracking.
Archived: Preserved but no longer deployable. Read-only. Terminal state.
| From | To | Allowed |
|---|---|---|
| Draft | Active | Yes — requires name, prompt, model, governance tier |
| Draft | Archived | Yes — abandoned draft cleanup |
| Active | Suspended | Yes — emergency pause or maintenance |
| Active | Archived | Yes — direct retirement |
| Suspended | Active | Yes — after issue resolution |
| Suspended | Archived | Yes — permanent retirement |
| Archived | any | No — terminal state |
Deployment States
Deployment is a separate dimension from the definition lifecycle. An active agent can be at a different deployment state on each gateway independently:undeployed → deploying → deployed | failedThe deployment matrix shows which agents are live on which gateways. Each deployment tracks sync status:
synced, pending_push, push_failed, or drifted (gateway config diverged from Convex source of truth).A2A Discovery
Active agents can be published for Agent-to-Agent (A2A) discovery via the AgentCard format. Publishing makes the agent discoverable by external agent systems, exposing its capabilities, endpoint, and supported protocols. This is opt-in and governed by the agent’s tier.Config Revision History
Every change to an agent definition is tracked inagentConfigRevisions with sequential revision numbers, field-level diffs, and change attribution. This supports audit compliance, rollback analysis, and change traceability.
In Burgundy: Manage lifecycle and deployments from the agent detail page. →