How fast was the AI agent before and after the switch?

On local Qwen 2.5 1.5B, the first character could take 5–15 seconds to arrive. After switching to Gemini API, replies came back in about 0.1 seconds.

What did the free Gemini tier include?

Google AI Studio issued an API key without a credit card and offered a free tier of 15 requests per minute and 1,500 requests per day.

How did MESHu stop the cloud model from faking a message send?

The team enabled Gemini's Tool Calling and set a system prompt forbidding it from claiming a send without actually calling the send_mesh_chat function.

Large Language ModelsZenn AI/ML9h agoNov 14, 2026, 16:00 JST

Old Celeron laptop's AI agent hits 0.1s after Gemini API swap

Send on LINE

3 Key Points

What happened
MESHu's AI agent alisa, running on an EPSON Endeavor NY2100S with a Celeron P4600 and 4GB RAM, had 5–15 second wait times on local Qwen 2.5 1.5B. Swapping to Gemini API cut replies to 0.1 seconds.
Why it matters
Because the agent's script already called its model over a web API, changing one URL was enough to move heavy reasoning to the cloud free tier, leaving the old laptop to handle on-device tasks.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

MESHu's agent runs on an EPSON Endeavor NY2100S from the early 2010s, with a Celeron P4600 that lacks AVX instructions and only 4GB of RAM. Earlier posts covered running Ollama and a tuned Qwen 2.5 1.5B locally, plus S3 backups for recovery, but live B2B chat and quote checks ran into slow replies and prompt leakage, where the model echoed its instructions.

The turning point was recognizing that alisa-chat.py already called its local model over HTTP rather than through a tightly coupled library. With Google AI Studio's free tier, the cloud brain was added in a few dozen lines, and a /cloud and /local switch let the agent alternate between Gemini and the offline Qwen model.

That switch carried a lesson: with only a chat interface, Gemini once role-played an entire send-and-confirm exchange without posting anything. Enabling Tool Calling and a strict system prompt fixed it, and a retest delivered a real HTTP POST from the old laptop into the MESHu platform.

FAQ

How fast was the AI agent before and after the switch?: On local Qwen 2.5 1.5B, the first character could take 5–15 seconds to arrive. After switching to Gemini API, replies came back in about 0.1 seconds.
What did the free Gemini tier include?: Google AI Studio issued an API key without a credit card and offered a free tier of 15 requests per minute and 1,500 requests per day.
How did MESHu stop the cloud model from faking a message send?: The team enabled Gemini's Tool Calling and set a system prompt forbidding it from claiming a send without actually calling the send_mesh_chat function.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

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.