Dashboard / Shared Infrastructure / Integration

Integration

External systems, LLM providers, networking, and channel bridges
Status: In progress WIP Runtime: OpenClaw on Mac Mini Networking: Tailscale Funnel (public HTTPS)
Architecture

The Integration layer connects the Understanding Workbench to external systems. It has three responsibilities: providing public HTTPS access to services running on the Mac Mini, connecting to LLM providers for AI capabilities, and bridging messaging channels so users can interact with the system from wherever they already work.

The design principle is that the Mac Mini is the brain — it runs the agent runtime (OpenClaw), hosts capability services (FastAPI backends), and manages API keys. The Vercel frontend is a thin presentation layer that routes to the Mac Mini for all substantive work.

┌─────────────────────────────────────────────────────┐ │ Public Internet │ │ │ │ vercel-app-psi-lime.vercel.app (Presentation) │ │ ├── Landing page / capability cards │ │ └── Guide chat → proxies to Mac Mini API │ │ │ │ Slack / WhatsApp / WebChat (Channel bridges) │ │ └── Route through OpenClaw Gateway │ └──────────────┬──────────────────────────────────────┘ │ HTTPS via Tailscale Funnel ▼ ┌─────────────────────────────────────────────────────┐ │ Mac Mini │ │ │ │ OpenClaw Daemon ──→ Anthropic API (Claude) │ │ ├── Guide Agent (soul.md + skills) │ │ ├── Capability skills (Library, Modeling, etc.) │ │ └── Gateway (ws://127.0.0.1:18789) │ │ │ │ FastAPI Services ──→ Python pipeline execution │ │ ├── Library Skills (:8000) │ │ └── Future capability backends │ │ │ │ API Keys via 1Password service account │ └─────────────────────────────────────────────────────┘
Components

OpenClaw

Open-source AI agent runtime. Runs as a launchd daemon on the Mac Mini. Provides the Gateway (WebSocket control plane), skill execution, MCP server support, and bridges to 25+ messaging channels.

agent runtime

Tailscale Funnel

Exposes local services to the public internet over HTTPS with a stable *.ts.net URL. No port forwarding, no dynamic DNS, no certificates to manage. Built into the OpenClaw ecosystem.

networking

Anthropic API

LLM provider for all AI capabilities. Claude models power the Guide agent, the Briefing skill's conversational interface, and LLM-assisted classification in the Enrichment skill.

llm provider

1Password Service Account

Secure credential management. API keys are stored in the bot Vault and accessed via OP_SERVICE_ACCOUNT_TOKEN. No secrets in code, no secrets in environment files.

credentials

Connection Map

FromToProtocolPurpose
Vercel frontendMac MiniHTTPS (Tailscale Funnel)Guide chat, capability UIs
OpenClaw GatewayAnthropic APIHTTPSLLM inference for all agents
OpenClaw GatewaySlack / WhatsApp / etc.WebSocket / webhooksChannel bridges for chat access
OpenClaw GatewayFastAPI serviceslocalhost HTTPSkill invocation (pipeline execution)
FastAPI servicesAnthropic APIHTTPSLLM calls for enrichment, briefing
OpenClaw Daemon1PasswordHTTPSAPI key retrieval
Channel Access

OpenClaw bridges messaging channels so the same Guide agent and capability skills are accessible from multiple surfaces. A user can start a conversation in the web UI and continue via Slack, or invoke a Library skill from WhatsApp.

Priority channels for initial deployment:

Channel Priority

ChannelStatusNotes
Web UI (Vercel)Live (direct API)Migrate to route through OpenClaw Gateway
OpenClaw WebChatAvailable with OpenClawBuilt-in, no additional config needed
SlackPlannedEnterprise pilot channel
WhatsAppPlannedCommunity/individual access
Setup Reference

The Mac Mini infrastructure setup is documented in the Library Skills app's PLAN.md (Phase 0). The same infrastructure serves all capabilities, not just the Library.

← Back to Dashboard