Skip to main content

​
Debug

Debug docs cover operator and admin diagnostics. For first setup, start in Agent > Models, Agent > Channels, Agent > Skills, Agent > Tools, Agent > Memory, Agent > Services, and Agent > Tasks. Use Debug when the setup page says runtime state does not match saved config, when a plugin is not loaded, or when you need raw evidence for support.

​
Where to Look

  • Advanced > Config: raw ~/.fased/fased.json escape hatch for settings that do not yet have a focused page.
  • Advanced > Debug: ordered diagnostics, status/health/model snapshots, task/run diagnostics, provider catalog checks, plugin runtime diagnostics, memory repair preview, event log, and raw RPC inspection.
  • Advanced > Nodes: paired devices, node capabilities, runtime clients, exposed command counts, pending pairings, and command exposure checks.
  • Logs: live gateway file logs, filters, auto-follow, export, and redacted operator troubleshooting.
  • Usage: local model usage history by Agent, provider, model, session, task, channel/source, tokens, cache tokens, and priced/unpriced cost.
First setup should stay on the focused Agent or runtime page. Use those pages for models, channels, skills, services, wallets, and tasks.

​
Common Workflows

Gateway looks offline
  1. Check the top-bar health dot.
  2. Open Logs and filter for startup, auth, provider, channel, or plugin errors.
  3. Open Advanced > Debug only when you need the full status/health snapshots or raw RPC result.
A channel, service, or skill says saved but not live
  1. Check the focused page first: Agent > Channels, Agent > Services, or Agent > Skills.
  2. Check Extensions if the surface comes from a plugin.
  3. Use Advanced > Debug for plugin runtime diagnostics, provider catalog checks, and raw status snapshots.
A node or device is paired but cannot execute commands
  1. Open Advanced > Nodes.
  2. Confirm the node is paired, recently connected, and advertising the expected capabilities.
  3. Check the command exposure counts before changing exec or sandbox policy.
Token usage looks wrong
  1. Open Usage.
  2. Filter by Agent first, then provider/model/session/task/source.
  3. Treat provider quota/account status separately from local token accounting. Local usage prefers transcript and task run-log usage records over session context snapshots.
Memory diagnostics need repair
  1. Use Agent > Memory for the selected Agent’s session-memory and QMD health.
  2. Use Advanced > Debug for Memory Doctor repair preview and gated repair execution.
  3. The preview is read-only. Repair execution requires the explicit admin path.
A task appears stuck or missing from Agent > Tasks
  1. Open Agent > Tasks first and check the Task row and latest run.
  2. Use cancel/retry from the row when the action is available.
  3. Open Advanced > Debug for stale-running, lost-task, and queue reconciliation warnings.
  4. Use Logs only when the run detail does not include enough failure detail.

​
Boundaries

  • Debug is evidence, not setup. If a friendly page exists, make the change there first.
  • Logs are redacted diagnostics. Still review before sharing externally.
  • Nodes are device/runtime surfaces. They are not normal chat channels.
  • Usage is local accounting. Provider quota, billing portals, and OAuth account limits are separate.
  • Task/run diagnostics are local runtime evidence. They are not full workflow orchestration; future Fased Workflows should build on stable run evidence.
  • Raw Config is the escape hatch. Do not use it for routine Agent setup.
See also: