Brickface AI Infrastructure — Deployment Dashboard

Service Health

OpenClaw Gateway v2026.3.24

Port 3000 — Self-healing (--force + min_uptime 10s)

Slack Socket Mode Cerebras Models LangGraph Routing Caddy SSL Tailscale

LangGraph Supervisor v3.0.0

Port 8080 — bind_tools + ToolNode + Cerebras + Groq

bind_tools tool_node cerebras groq-tool-calling vault marketing-delegation multi-model

LangGraph Agents v0.6.0

Port 8100 — 20 agents, 3 domains, MCP tools, Cerebras

multi-model mcp-tools cerebras subgraphs auto-routing

Agent Domains (20 Agents)

Marketing 12 agents

reporting gbp lead-scoring campaign-lifecycle meta-ads bing-ads creative-pipeline weather-campaigns review-collection content-pipeline nurture offline-conversions

Finance 4 agents

ar-aging cash-flow expense-analysis pnl-analysis

Tools: QBO, Mercury, Ramp via MCP Gateway

Engineering 4 agents

issue-triage code-review infra-ops cursor-dispatch

Tools: GitHub, Cursor Cloud (256 max), PM2

LLM Providers

Cerebras llama3.1-8b

6,728 tok/s

Cerebras qwen-3-235b (PRIMARY)

1,207 tok/s

Groq llama-3.3-70b (TOOL CALLING)

335 tok/s

Gemini 2.5 Flash (1M CONTEXT)

79 tok/s

Claude Opus 4.6 (REASONING)

Strategy Only

Bound Tools (Supervisor v3.0)

search_hubspot_contacts

search_vault

call_marketing_agent

scan_github_issues

create_clickup_task

query_quickbooks

check_mercury_balance

LLM decides WHEN to call tools via Groq bind_tools(). 3-call loop cap prevents runaway. Cerebras fallback on Groq failure.

Connected Services (370 MCP Tools)

HubSpot

ClickUp

QuickBooks

Mercury

Ramp

Gmail

Google Drive

Google Calendar

n8n Workflows

GitHub Projects

Obsidian Vault

Cursor Cloud

Slack

Dialpad

Canva

Figma

Cloudflare

Linear

Autonomous Cron Jobs

*/30 * * * * orchestrator-scan Scans GitHub Projects, dispatches Cursor agents on top 3

0 8 * * * ET daily-marketing-report Google Ads spend, leads, GBP reviews, content pipeline

*/2 * * * * agent-chat Agent heartbeat and chat monitoring

*/15 * * * * brain-sync Obsidian vault synchronization

*/15 * * * * swarm-security Security scan of open P0/urgent issues

*/30 * * * * swarm-engineering Engineering issue processing

0 * * * * swarm-growth Marketing/growth issue processing

*/5 * * * * infra-builder-nonstop Infrastructure build monitoring

*/2 * * * * website-sprint Website rebuild sprint tracking

Tailscale Path Routing

/ :3000 OpenClaw Gateway

/health :3001 MCP Gateway

/sse :3001 MCP Gateway (SSE)

/mcp :3001 MCP Gateway (Streamable HTTP)

/langgraph :8080 LangGraph Supervisor

/lg-agents :8100 LangGraph Agents

:8443 :3004 OpenClaw Studio

:8444 :8080 LangGraph Supervisor (direct)

:8445 :8100 LangGraph Agents (direct)

Deployment Timeline

Crash Loop Identified

OpenClaw gateway at 2,358 restarts. Orphan PID 2132291 holding port 3000.

Self-Healing Fix Deployed

start-gateway.sh rewritten with --force flag. ecosystem.config.js updated with min_uptime: 10s. Crash loop eliminated permanently.

LangGraph Supervisor v2.0

Cerebras multi-model routing. Obsidian vault tools. Marketing agent delegation. 5-node graph.

12 Agents Upgraded to v0.5.0

MCP tool access (HubSpot, ClickUp, Vault). Cerebras primary LLM. Auto-routing /route endpoint.

Cron Jobs Registered

orchestrator-scan (*/30min) + daily-marketing-report (8am ET). 9 total cron jobs.

Supervisor Upgraded to v3.0.0

Real bind_tools() with 7 LangChain tools. Groq handles tool calling, Cerebras for reasoning. 3-call loop cap.

Finance + Engineering Subgraphs

8 new agents: AR aging, cash flow, expense analysis, P&L, issue triage, code review, infra ops, Cursor dispatch. Total: 20 agents.

Tailscale Path Routing Fixed

/health, /langgraph, /lg-agents, /sse, /mcp all route to correct backends. 502 eliminated.

Before vs After

BEFORE (Start of Session)

LangGraph: ~15%

Agents: 12 (prompt-only)

Tool access: None

LLM providers: Groq only

OpenClaw status: Crash loop (2,358)

Supervisor: v1 (2 agents)

Finance agents: 0

Engineering agents: 0

Autonomous crons: 7 (no orchestrator)

Tailscale /health: 502

AFTER (End of Session)

LangGraph: ~90%

Agents: 20 (3 domains)

Tool access: 7 bind_tools + 370 MCP

LLM providers: 5 (Cerebras primary)

OpenClaw status: Self-healing (0 restarts)

Supervisor: v3.0.0 (5 agents + tools)

Finance agents: 4 (AR, cash, expense, P&L)

Engineering agents: 4 (triage, review, ops, Cursor)

Autonomous crons: 9 (+ orchestrator + marketing)

Tailscale /health: 200 OK