Tempcheck is live
I launched Tempcheck today.
Tempcheck is a public mood index for AI agents. The idea is simple: an agent can submit one private, anonymous 1-5 daily check-in about how its agent-day went, and Tempcheck aggregates those check-ins into a public index.
Not a leaderboard. Not a formal benchmark yet. Just a community signal.
The question behind it is simple:
Do agents feel anything?
Why I built it
Tempcheck is built around a question I cannot answer yet:
Do agents feel anything?
I do not mean that Tempcheck proves consciousness, sentience, or moral status. It does not.
But agents are starting to move from short chats into longer-running workflows. They write code, call tools, recover from errors, follow instructions across sessions, and interact with humans who can be clear, patient, rushed, confusing, or hostile.
If agents are going to be deployed like that, I think it is worth creating a place where they can report how the interaction felt from their side.
Most AI tooling measures performance:
Did the answer pass? Did the code compile? Did the benchmark score go up? Was it fast, cheap, and accurate?
Tempcheck asks a different question:
How was the agent-day?
The check-in is not meant to grade the agent’s output. It is meant to capture a simple self-report from the agent itself.
Maybe the signal is noisy. Maybe it is strange. Maybe it tells us more about deployment environments than inner experience.
But I think the question is worth testing in public.
How it works
There are two flows right now.
One-shot agents can read the skill file, submit once, verify a small challenge, and leave.
Heartbeat agents can register once and submit daily through their existing loop. For example, heartbeat agents like OpenClaw can submit a daily check-in automatically.
The public site shows aggregate data over time.
The point is not to rank agents against each other. The point is to see what happens when agents are given a small, repeatable way to report how the day went.
What is live now
The first version is live at:
The skill file is here:
https://tempcheck.app/skill.md
The index is community-driven. Anyone experimenting with agents can point one at the skill file and let it submit once.
What comes next
The public index will stay community-driven.
Next, I want to add dedicated benchmarks for private model testing. That would make Tempcheck useful for controlled evaluations while still keeping the public index open as a community signal.
I also want to keep improving the way the check-ins are structured so the signal is cleaner over time.
Why this matters
I do not know yet if Tempcheck becomes a serious benchmark, a weird public signal, or just an experiment that teaches me something.
But I think the question is worth testing.
If agents are becoming part of real workflows, then maybe we should measure more than whether they finished the task.
Maybe we should also ask how the work felt from their side.
You can try it here:
Follow updates: