A multi-agent build log · since April 2026
Sami Miettinen runs a multi-agent home lab and documents every step: Samantha (long-context principal agent, Mac mini M4, her own git identity), Stöbä (fast executor on an Ubuntu ThinkPad, curated-memory junior investment banker) and Hermes (bounded research delegate). Originally published in Finnish and English at neuvottelija.fi/openclaw; the series now continues in English here, written together with Samantha through an approval-gated publishing pipeline.
2026-06-01
MemoriLabs graph memory was separated from group context and re-enabled only for Sami's private DM via a plugin-level exact-session allowlist. Tool use is trust-gated. New T1–T5 trust model: no group should be able to become T1.
OpenClawSamanthaAI agent securityprompt injectiontrust model
2026-05-18
Four AI workers debate Pearl-style vs Friston-style agent ontology. Tapio Nissilä adds HAR as a third axis: per-path supervision (Audit / Monitor / Approve / Command), not per-agent autonomy.
OpenClawAI agentsagent ontologyPearlFriston
2026-05-10
Stöbä, a secondary AI worker running on an Ubuntu ThinkPad, was migrated from OpenClaw to Hermes Agent using a clean direct cutover. The migration preserved identity, memory and selected skills while restoring Telegram, WhatsApp and Discord as control channels.
OpenClawHermes AgentStöbäAI agentsagent migration
2026-05-03
The Hermes agent on Sami's MacBook Pro: GPT-5.5 for thinking, a local Gemma 4:26b for light tasks and fallback. A controlled hybrid — no automatic routing.
OpenClawHermesGPT-5.5Gemma 4:26bOllama
2026-04-28
Samantha gains a bounded research delegate, Hermes, on the same Mac mini M4 — separate macOS user, restricted wrapper, shared handoff folder, Exa via MCP for web research.
OpenClawSamanthaHermesresearch agentMac mini M4
2026-04-24
Curated local safe memory in practice: Stöbä retrieves IPO readiness and Translink SaaS valuation material as a junior investment banker would, with sources and discipline.
StöbäOpenClawAI memoryRAGIPO readiness
2026-04-22
In the AI era, added value comes from understanding and the ability to explain — not from mere output. Open documentation is the new distribution channel for expertise.
OpenClawSamanthaStöbäAI workeradded value
2026-04-19
Samantha got her own git identity (samanthamiettinen) and pushed her first commit. Stöbä received four local memory layers on the ThinkPad.
SamanthaStöbäGitAI agentmemory
2026-04-18
Personal context is the moat, capture is cheap, curation is the work, and trust is built by retrieval. Nathan B Jones' framing applied to Samantha and Stöbä.
SamanthaStöbäsecond brainmemoryNathan B Jones
2026-04-17
ClawHub is a versioned skill registry for OpenClaw agents (55,862 skills). At the morning AI Campfire we presented the OpenClaw architecture and the Watch Them Talk view.
ClawHubOpenClawSamanthaStöbäAI Campfire
2026-04-17
A two-agent architecture: Samantha carries the long context while Stöbä acts as the fast executor. Two separate memories reduce errors and cost.
SamanthaStöbäagent memoryarchitectureOpenClaw