CoachB OS — Settings

Agent Configuration

Model assignments and heartbeat settings for all 21 agents. Changes require an OpenClaw node restart.

groupModel Assignments

Agent	Role	Assigned Model	Heartbeat

monitor_heartHeartbeat Settings

Stall Detection Threshold

Agents flagged red after this many minutes without a heartbeat

Auto-refresh Interval

How often the dashboard polls for agent updates (seconds)

Telegram — Bobby Bot

Bot username for direct Bobby communication

Telegram — Sid Bot

Bot username for Sid personal assistant

Workflow Rules

Gate configuration, WIP limits, and approval routing. These are the rules the system enforces automatically.

ruleHard Rules — Never Override

Nothing publishes without Coach B approval

Hard Gate 4 requires explicit trigger — no automatic publishing ever

Nothing spends money without Coach B approval

All payment actions require Level 4 authorization

Joe veto authority always active

Joe can block any deployment, skill install, or content release

Bayo FLAG FOR COACH rule

Any content involving Bayo requires personal Coach B review — cannot be delegated

stacked_bar_chartWIP Limits

Maximum Episodes in Pipeline

Total WIP across all 15 stages simultaneously (Little's Law — keep at 3-4)

Max Episodes at Any Single Gate

Coach B reviews one item at a time — prevents approval fatigue

Review Buffer Size

Items pre-validated and ready for Coach B before new work starts

account_treeTributary Soft Unlock Rules

Localization glossary prep unlocks at

When Elena can begin language work

Marketing hook drafts unlock at

When Sofia can begin thumbnail and social copy

Shorts extraction unlocks at

When clip shorts can be pulled

File Paths

Canonical folder locations used by all agents and the dashboard. Change with caution — agents read these paths from their memory files.

folder_openSystem Paths

CoachB OS Base

Root folder for all agent files, skills, and outputs

~/Documents/CoachB_OS/

Kids Club — World Bible (canonical)

Official living source of truth — all agents read from here

~/Documents/KidsClub/WorldBible/

Staging Folder

Where agents drop files for Coach B review

~/Documents/CoachB_OS/Staging/

Agent Skills

Per-agent SKILL.md files — loaded automatically at session start

~/Documents/CoachB_OS/Agents/<Name>/skills/

OpenClaw Config

Main config file — only file OpenClaw reads

~/.openclaw/openclaw.json

Shared Skill

coachb-self-improving — installed for all agents

~/.openclaw/skills/coachb-self-improving/

Notifications

Control how and when the system alerts you. Telegram is the primary channel for agent communication.

notificationsNotification Channels

Telegram alerts from Bobby

Bobby sends briefings, gate requests, and escalations via Telegram

Dashboard review queue badge

Show pending count on the Agents tab when items need review

Stalled agent alerts

Notify when an agent has not sent a heartbeat for 20+ minutes

Security flag notifications

Immediate alert when Joe raises a kid-safety or security flag

scheduleReview Windows

Tuesday Review Window

Creative review — scripts, story, characters

Thursday Review Window

Production review — visuals, localization, final packages

System

Ollama configuration, memory budgets, and server settings.

memoryOllama Settings

OLLAMA_MAX_LOADED_MODELS

Max models in memory simultaneously — prevent RAM overcommit

OLLAMA_KEEP_ALIVE (hot models)

How long to keep frequently-used models loaded (minutes, -1 = infinite)

OLLAMA_KV_CACHE_TYPE

Set to q8_0 to halve KV cache memory with negligible quality loss

RAM Headroom Reserve

Minimum free RAM to maintain — system warns if exceeded (%)

lanServer Ports

Dashboard (Vite)

Front-end dev server

Status Server

Agent heartbeats and Control Room backend

OpenClaw Gateway

Agent communication hub

OPS Dashboard (legacy)

radio_tower.py — original dashboard

Danger Zone

These actions are destructive or irreversible. Read carefully before proceeding.

⚠ Safe Restart (use this, not openclaw reset)

Stops and restarts the OpenClaw node without wiping Telegram credentials or session data. Use this when agents are unresponsive. Do NOT use openclaw reset — it destroys credentials.

⚠ Clear Review Queue

Removes all items from the staging review queue. Does not delete files — only clears the queue database. Use if the queue is corrupted or you want a fresh start.

🚫 Never Use openclaw reset

openclaw reset is destructive — it wipes Telegram credentials, session data, and agent configurations. It cannot be undone. If you accidentally run it, you will need to re-authorize all Telegram bots and re-configure all agents. Always use the safe restart sequence above instead.