Turn it up
All projects

Project detail · technical deep-dive

LLM orchestrationAI agentsTool streamingMulti-channel AIOpenClawTypeScriptNode.jsDockerWebSocket

Friday — AI Orchestration & Agentic Infrastructure

Architected and deployed a production-grade autonomous agent stack using OpenClaw: a local-first WebSocket control plane, Dockerized orchestration, and multi-channel productivity integrations (WhatsApp, Slack, Discord).

Role

Solo / systems engineering

Timeline

2026

Tech

Node 22+, TypeScript, Docker Compose, OpenClaw CLI, Vitest

9 page views

Product showcase

Flows & screens

Gateway and product surface: local-first control plane, agent runtime, and skills wired into real workflows.

Active setup

Status

Agents & ops

Focus

Node 22+

Runtime

Technical implementation

  • Gateway & control plane

    Local-first WebSocket gateway for session material, auth, and routing. Built for reliable LLM connectivity and explicit security policies.

  • Agentic execution

    Pi-style agent runtime with tool streaming and task handling beyond simple prompt-response, composed with OpenClaw primitives.

  • Multi-channel routing

    Inbound pairing, allowlists, and DM routing across messaging APIs, with hygiene via doctor checks and pairing approvals.

Overview

Friday is a high-agency AI orchestration system I built to manage complex workflows. Leveraging the OpenClaw stack (TypeScript / Node.js), I implemented a local-first gateway that acts as a centralized WebSocket control plane.

More than just a chatbot, this project is an exercise in systems engineering: managing modular agent lifecycles, complex channel routing, and maintaining reliable auth and failover patterns for LLM connectivity. It connects daily interfaces (WhatsApp, Slack, Discord) to a persistent, tool-enabled assistant.

Technical architecture & ownership

  • Full-stack engineering: Developed on Node.js 22+ and TypeScript, utilizing pnpm for workspace management and performance.
  • Infrastructure & DevOps: Orchestrated via Docker Compose, ensuring a reproducible environment for the gateway and CLI. Persistent state through volume-mounted workspaces and environment-driven secret management.
  • Agentic workflows: Implemented a Pi-style agent runtime capable of tool streaming and task handling, moving beyond simple prompt-response into autonomous execution.
  • Multi-channel routing: Unified gateway handling inbound message pairing, allowlist security policies, and explicit DM routing across messaging APIs (grammY, discord.js, Baileys).

Stack

  • Runtime: Node.js 22+, TypeScript
  • Orchestration: Docker & Docker Compose
  • Communication: WebSocket (control plane), webhooks, OAuth
  • Security: DM pairing, token-based gateway auth, environment secret management
  • Tooling: OpenClaw CLI, Vitest for CI-style checks

System operations (CLI)

Key commands for maintenance and execution:

  • openclaw gateway — persistent control plane and session material
  • openclaw agent — agent runtime with higher reasoning depth for complex tasks
  • openclaw doctor — hygiene, migrations, and risky DM policy audits
  • openclaw pairing — security-first approval for unknown DM senders

Engineering highlights

  • Fast deployment: From zero to a functional agentic system by composing and customizing open-source core modules.
  • Operations-minded: Explicit allowlists and “doctor” checks so the assistant stays safe in real messaging environments.
  • High-agency setup: A skills layer extends behavior through workspace skills, including automation into the Notion API.