AI Agents
Coming soon
A full agents case study is on the way. Here is what exists today — the OpenClaw-style remote agent I am building now, the Jarvis local stack that came before it, and the desk agents that keep the machine running.
What is landing here
This pillar will document the full agent architecture — setup, boundaries, memory, channels, and how each surface stays in its lane. Until that write-up ships, the stack below is the honest snapshot.
OpenClaw — SuperGrok 9000
In active buildPhone-first remote agent I am building now — Telegram channel, Grok 4.3 (xAI OAuth) as the brain, OpenClaw gateway on localhost, headless browser for web tasks, local memory index, and skills like PlumbForge (natural-language image search from the vault back to Telegram).
- ·Owner-paired Telegram — remote ops when away from the desk
- ·Gateway + browser as user-level systemd services on Ubuntu
- ·Intentionally separate from desk Claude and local Ollama
- ·Health and ops surfaced in AI Command Center today
Jarvis
Previous local agentThe local agent stack before OpenClaw took over remote work — API daemon with tool-use REPL, persistent sub-agents (coder, vision, security), and pairing with Ollama + Open WebUI on this machine.
- ·26 tools — local execution, not cloud-dependent
- ·Launched via `jarvis chat` / `ai chat` on the workstation
- ·Still running for local REPL and specialist sub-agents
- ·Day-to-day remote automation moving to OpenClaw
Desk agents
Operating todayThe builders that wire everything together — Grok terminal for repos and system work, Claude Code as parallel desk integrator, Grok browser for creative and strategy.
- ·Grok Build — context files, site work, OpenClaw ops
- ·Claude Code — dashboard UI, integrations, specialist subagents
- ·Clear provider boundaries by design
OpenClaw architecture (preview)
Telegram (owner-paired) → openclaw-gateway (127.0.0.1:18789) → Grok 4.3 (xAI OAuth) → headless browser, memory index, skills → PlumbForge: natural language → images on Telegram
See how this fits the wider machine in the AI Command Center case study.
Questions about this project?
AI Command Center