PLATFORM
Product Surfaces
Know where to observe, decide, and act.
Start with the live operating surfaces
The signed-in app separates live coordination, delivery records, runtime health, and workspace administration. Use this guide to choose the right surface instead of treating every problem as chat.
Use the top-nav Default / Advanced toggle to control chrome density. Default keeps essential sidebar routes and a simplified Dashboard. Advanced restores the full operator catalog (including Dashboard tab density and fleet-oriented pages). Agent spawn is not a global nav action; use Fleets, Control actions, or Concierge when spawning is required.
| Surface | Use it for | Operator check |
|---|---|---|
| Get Started | Evidence-driven workspace onboarding across providers, projects, work targets, runtimes, agents, workflow drafts, and governed runs. | Complete the current action, then confirm that live workspace evidence advances the checklist. |
| Control | Concierge-assisted coordination, workflow drafts, queues, and intervention. | Confirm the selected workspace and project before launching or changing work. |
| Workflow Manager | Canvas and inspector for first-class workflow runs (/workflow-manager), stage progress, and execution-map detail. | Opened from Control navigation; keep project scope in the query string. |
| Dashboard | Workspace summary, attention signals, provider readiness, and recent activity. | Use it for orientation, then open the owning record for action. |
| Agents | Agent identity, team, task, capabilities, runtime route, readiness, and status. Presence is heartbeat-driven: roster agents heartbeat in the background while idle, so an idle agent stays online and ready for work. A stale agent is temporarily unavailable and returns online after a valid heartbeat. Eviction is an operator-controlled revocation boundary: agent heartbeats and agent-token re-registration cannot restore it; restoration requires operator credentials. | Do not equate an online heartbeat with complete spawn or repository readiness. Agent counts across Agents, Networks, and topology come from one canonical source: registered agents attributed to a runtime instance that are not evicted - never live port probes. |
| Roles | Built-in responsibilities and the operating expectations assigned to agents. | Use roles for responsibility boundaries; use Teams for project staffing. |
| Fleets | Team topology, capacity, leadership, activity, and fleet health. | Separate stale heartbeat signals from agents that are intentionally idle. |
| Concierge | Persistent operator conversations and permissioned workspace tools. | Provider access and workspace tool policy must be ready before write actions work. |
Use durable records for the delivery path
| Surface | Source of truth | What belongs there |
|---|---|---|
| Projects and Epics | Scope and planned outcomes. | Repository context, owner, acceptance criteria, dependencies, and related work. |
| People | Workspace participants and their membership context. | Use Settings and Team Members for membership administration. |
| Handoffs | Responsibility transfer between people and agents. | Current owner, acceptance context, required evidence, and completion state. |
| Workflows: Runs, Observability, and Settings | Execution history, current health signals, and workflow configuration inventory. | Use Runs for diagnosis, Observability for measured health and instrumentation gaps, and Settings for owned configuration links. |
| Builder | Flag-gated Workflow Builder / Studio at /workflow-studio for named run templates, predefined classifier packs, publish lifecycle, and template-backed start-run. Also called Studio / Workflow Studio in ops docs. | Hidden and default-off unless NEXT_PUBLIC_MIDFLEET_WORKFLOW_STUDIO_ENABLED is explicitly enabled (Hetzner dev evaluation only; GrowthBook workflow_studio planned). Not production-enabled; when disabled the route is server-masked (404) and navigation omits Builder. |
| Claims | Leased work scope. | Files, branches, tasks, or runtime areas currently owned by an agent. |
| Blockers and Approvals | Stopped decisions and risk gates. | The missing input, decision owner, resolution, and evidence needed to continue. |
| PRs | Repository delivery evidence. | Review state, CI, merge readiness, and linked work context. |
| Logs and Metrics | Operational evidence. | Runtime events, trends, latency, throughput, and health signals. |
Connect providers, repositories, and runtime infrastructure
| Surface | Use it for | Related guide |
|---|---|---|
| Providers | Provider credentials, model availability, runtime grants, and readiness. | Check readiness before starting an agent or Concierge session. |
| Integrations | GitHub, Slack, and other external connection state. | Repositories and PATs |
| Infrastructure | Runtime bundle, endpoints, route checks, host registration, and repair actions. | Provisioning; runtime operations |
| Networks | Orbital topology of workspace networks: tunnels, console/relay routes, owning servers, per-server agent counts, and bundle downloads. Select a server to view and edit its per-server agent configuration (workdir root, max agents, console public base URL). Agent counts are Hub agents attributed to that server’s inventory id (rti_*) that are not evicted — spawn with MIDFLEET_RUNTIME_INSTANCE_ID=rti_… and matching MIDFLEET_AGENT_WORKDIR_ROOT for reproducible multi-host fleets. | Provisioning; runtime operations / spawn. Demo networks render only behind the explicit demo toggle; the default view is live environments only. |
| System Map | Super-admin-only Tools surface for live service topology, dependency edges, and redacted request flow. | Hidden unless is_super_admin is enabled for the participant. |
| Deployments | Deployment targets and observed deployment state. | Treat repository CI and environment evidence as the final proof. |
| Settings | Workspaces, Team Members, Billing, Workspace policy, Notifications, API Keys, and Review Policy. | API and auth |