Skip to content

Runtime Diagnostics

Runtime Diagnostics is the first place to check when prompts fail, capabilities look missing, or UI behavior degrades.

In VS Code, open Runtime Dashboard → Diagnostics and choose Refresh diagnostics. On desktop, use the affected panel’s connection status and Help → Activity → Ask Sophia for a recorded task issue. Available fields vary by host.

Core Fields

Typical diagnostics include:

  • workspace path
  • session state and active mode
  • selected model and agent
  • connected providers and model counts
  • permissions mode
  • capability health states

Capability States

Each capability reports:

  • AVAILABLE
  • PARTIAL
  • UNAVAILABLE

and a source (for example server, file, or hybrid).

Fast Triage Sequence

  1. refresh diagnostics
  2. verify /runtime/health
  3. verify /runtime/capabilities
  4. confirm provider auth and model visibility
  5. retry the prompt

CLI Health Checks

These examples apply to a manually started runtime at the shown address. Use your configured host, port, and required authentication. The packaged desktop manages its embedded runtime and can use a different port; use the app’s connection information rather than assuming 4096.

Terminal window
curl http://127.0.0.1:4096/runtime/health
curl http://127.0.0.1:4096/runtime/capabilities

Stuck Or Looping Task Diagnostics

Use alphabase doctor together with /runtime/health and /runtime/capabilities for ordinary client triage. Persistent-runner control and queue diagnostics are operator-only deployment details and are intentionally not part of the public client contract. Do not expose debug or control endpoints to untrusted clients.

Common Diagnosis Patterns

UNAVAILABLE + server

Usually runtime auth mismatch, server outage, or bad target host/port.

PARTIAL + hybrid

Fallback data path is active; live capability fetch is degraded.

Local AI unavailable

Local runtime (for example Ollama) is unreachable or not started.

Escalation Packet

When escalating:

  1. app surface (Desktop/VS Code/TUI/Web)
  2. exact prompt used
  3. diagnostics snapshot
  4. runtime logs in failure window
  5. screenshot of final UI state

Review this material before sharing it. Remove API keys, passwords, access tokens, authorization headers, private file contents, personal paths, and confidential prompt text. Share a minimal reproduction and relevant error excerpts rather than an entire workspace or log archive. Do not assume screenshots or diagnostics exports are automatically redacted.