You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
fix: add WebSocket ping keepalive for long inference times
WebSocket connections silently drop during long Ollama inference
(8+ min on CPU) because no data flows and proxies/NAT close idle
connections. Add periodic ping frames every 30s to keep alive.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
fix: pull default Ollama model when leader uses inherit
When the leader's SubAgentModel is "inherit", the deploy skipped the
Ollama model pull but OPENCODE_MODEL was set to the default model
(e.g. qwen3:4b). This caused "model not found" errors at runtime.
Now falls back to defaultOpenCodeModel() for the pull step.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>