Existing proposals
Problem
A single-process development runtime is insufficient for multiple repositories, concurrent tasks, durable execution history, and isolated remote runners. Agent Zero needs a production control plane while keeping execution providers outside the core agent runtime.
Parent roadmap: #5
Desired outcome
Deliver the v0.3 control-plane and runner-pool milestone.
Acceptance goals:
- productionize the Nitro-based control plane
- add persistent task history and structured event logs
- add cost, token, latency, and model-usage tracking
- introduce provider-neutral remote sandbox orchestration
- evaluate ViteHub Sandbox / Workspace / Box adapters for Cloudflare, Vercel, and other providers
- add runner pools, scheduling, quotas, and lifecycle management
- add a dashboard for task state, findings, verification evidence, and approvals
Likely area
oRPC server
The roadmap is moving this boundary toward Nitro v3; the feature template still names the existing server area as oRPC server.
Safety and compatibility
High impact. Remote execution must remain behind packages/runner; provider credentials must not leak into agent state or logs; task persistence must avoid storing secrets; quotas and lifecycle controls must prevent abandoned or unbounded sandboxes. The control plane must not become a second command-execution path.
Alternatives considered
Embedding Cloudflare, Vercel, or another provider SDK directly into the agent runtime was rejected because it would couple orchestration to infrastructure. A provider-neutral runner/sandbox boundary is preferred. Keeping the control plane ephemeral was rejected because durable task history and concurrent scheduling require persistence.
Existing proposals
Problem
A single-process development runtime is insufficient for multiple repositories, concurrent tasks, durable execution history, and isolated remote runners. Agent Zero needs a production control plane while keeping execution providers outside the core agent runtime.
Parent roadmap: #5
Desired outcome
Deliver the v0.3 control-plane and runner-pool milestone.
Acceptance goals:
Likely area
oRPC server
The roadmap is moving this boundary toward Nitro v3; the feature template still names the existing server area as oRPC server.
Safety and compatibility
High impact. Remote execution must remain behind
packages/runner; provider credentials must not leak into agent state or logs; task persistence must avoid storing secrets; quotas and lifecycle controls must prevent abandoned or unbounded sandboxes. The control plane must not become a second command-execution path.Alternatives considered
Embedding Cloudflare, Vercel, or another provider SDK directly into the agent runtime was rejected because it would couple orchestration to infrastructure. A provider-neutral runner/sandbox boundary is preferred. Keeping the control plane ephemeral was rejected because durable task history and concurrent scheduling require persistence.