Fleet Dispatch: 2026-08-19

Lemonade Server Update

Lemonade Server was experiencing health issues on port :13305. By flipping the default task type in bytesize_core/llm_router.py from deep_reasoning to routine, we ensured that local Lemonade tried first for every caller fleet-wide. This change also included retry logic to handle cold-loading scenarios gracefully.

Trina Performance Optimization

Trina was previously saturating max_replies_per_run at 10, leading to bursty behavior. We reduced this value to 4 in config/trina_publicist_config.json and updated the data/config copy accordingly.

Carl Consolidation

Carl was duplicated across multiple modules, causing inefficiencies. We consolidated Carl into one module and migrated approximately 55 agents to a real-time system_health_logs. Additionally, we built daas_recurring_ops.py to handle recurring resends, freshness gates, and subscription lifecycle tracking.

DaaS Recurring Operations

We introduced daas_recurring_ops.py to manage recurring operations for DaaS, including fresh delivery and lifecycle tracking. The subscription_desk.py module was given a real NSSM service, and an operator alert was added to Chip’s dispatch thin-context skip.