Root cause: Workers hung because default model was DeepSeek v4-pro (reasoning
model that produces empty content with normal token limits). pi RPC workers
waited forever for content that never arrived.
Changes:
- Circuit breaker (CLOSED→OPEN→HALF_OPEN) around Zulip API calls
- Retry with exponential backoff + jitter for transient errors
- Queue lifecycle management (idle_queue_timeout, auto re-register on BAD_EVENT_QUEUE_ID)
- External supervisor watchdog (restarts router after 3 health check failures)
- Busy worker timeout (SIGKILL after 5 min + error DM)
- PM2 hardening (max_restarts=100, max_memory_restart=500M)
- Crash handlers (uncaughtException + unhandledRejection → reconnect, not die)
- Fixed default model: deepseek-v4-pro → syslog-harness/syslog-auto
- Enhanced health endpoint with circuit breaker stats
Architecture: Research-backed from Zulip event system docs + Node.js resilience
patterns. Inline circuit breaker (no dependency). Separate watchdog process (Hermes pattern).
Verified: Circuit breaker trips on outage, recovers gracefully. End-to-end DM
processed in 4 seconds. Watchdog monitoring every 30s.
Complete rewrite of the pi Zulip extension as a standalone Node.js service
with direct harness API calls. No longer depends on pi's session — survives
pi shutdown and restart.
Changes:
- src/index.ts: Complete rewrite — standalone process, not a pi extension
- Direct harness API calls (POST /v1/chat/completions) with conversation memory
- Per-sender conversation history (up to 50 messages)
- Placeholder->edit streaming for Zulip DMs
- Typing indicators via Zulip API
- Health endpoint on :9200
- Exponential backoff reconnection
- Graceful shutdown on SIGINT/SIGTERM
- abiba-zulip.service: systemd service unit file
- README.md: Updated deployment instructions
- package.json/tsconfig.json: Updated for standalone app
Deploy: npm install -> npx tsc -> systemctl enable abiba-zulip
Closes issue #24 (Hermes parity for pi)
- Adds message_update handler that periodically edits the Zulip
placeholder with partial response text (every ~2s during streaming)
- Sends a 'Thinking...' placeholder to Zulip immediately on DM receipt
- Detects agent busy state and uses deliverAs: 'steer' so subsequent
Zulip DMs steer the conversation mid-stream (like terminal TUI)
- Adds editMessage() to the queue helper (Zulip PATCH /messages/{id})
- sendMessage() now returns the message ID for editing
- Turn_start / agent_end lifecycle tracks agent busy state
- Cleanup: streaming timer properly cleared on shutdown
- Removed spawnSync subprocess (pi -p --mode json) for Zulip DM processing
- Messages now inject directly into the current pi session via
pi.sendUserMessage(), reusing the already-hot LLM — zero cold-start
- agent_end event handler captures assistant response and relays it to Zulip
- Added pending reply queue for ordered delivery of multiple Zulip messages
- Fixed zulip-js import (CJS default export, not named createClient)
- Added types.d.ts for zulip-js with default export declaration
- Updated package.json with yaml dependency, @earendil-works/pi-coding-agent
- Config supports both nested YAML and flat env-var formats
- Removed all subprocess timeout/buffer error paths (no longer needed)
Abiba (pi)
jerome