
What happened
The author installed the open-source agent platform OpenClaw on an EPSON Endeavor NY2100S (Celeron P4600, 4GB RAM), removed the local Ollama model, and ran the agent alisa on Google Gemini Flash Lite's free tier, completing a MESHu quote-to-order deal worth ¥34,595.
Why it matters
It suggests a working autonomous procurement agent can run with no GPU server, paid cloud instance or paid API quota, which may lower the cost floor for firms experimenting with back-office AI, though this is a small single-case trial.
Summaries like this, every morning.
The write-up is the latest entry in a running series in which the author has been rebuilding an old laptop as a "field AI employee." Earlier installments replaced a slow local Qwen model with Google Gemini API's free tier and had alisa complete a mock B2B sales negotiation with tool calling. The remaining gap was persistence: alisa could only act while the chat script was running.
The fix came in two parts. Removing Ollama freed about 1.7GB of memory and brought swap use to zero, and loginctl enable-linger let the OpenClaw Gateway start on power-up without a login. The author also bound the agent's workspace into a Samba share, so its SOUL.md personality file could be edited from another machine.
Notably, the design deliberately avoids full automation. A rule in AGENTS.md makes alisa stop and ask which company to log in as when an instruction is ambiguous, and in a test it did exactly that rather than guessing between the seller and buyer roles. In the recorded scenario alisa acted as buyer-side procurement, filing the quote request, reading the seller's reply, and placing the order through MESHu. The author says the code is published on GitHub as MESHu-OSS.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
IMPACTLAKE began offering a ChatGPT plugin for impactlake(TM) and added Model Context Protocol (MCP) support…

A developer using Claude Code on a personal web app discovered via --debug logs that his Stop hook passed ever…

Internal documents seen by Reuters show Amazon plans a "World Product Graph" catalog of global product data, i…

株式会社粋農 implemented MCP (Model Context Protocol, a common standard linking AI apps to external data and functio…

AIworker published a free November 2026 report on structuring Microsoft 365 Copilot training for company-wide…

Anthropic struck a rare megadeal in AI infrastructure, committing $3.4 billion to take over orders for 1.7GW o…
