中文 日本語
home / what-is-kanonagent

# What is KanonAgent?

KanonAgent is an early radar for real traction signals in AI agents. Not a directory, not a "we index everything" database, and not a recommendation site — one job: which AI agents have crossed a real growth threshold, when that moment happened, what they were then and what they are now.

Once a breakout is written it is never rewritten — an append-only constraint at the database level, not a promise. Wrong calls stay up, because a radar that quietly edits its own history is worthless. See the methodology for the exact thresholds.

We chase no launch-day noise, no paid placement, no gut-feel picks — traction only.

Free, no login, no ads. English and Chinese.

What the name means

Kanon (κανών) is Greek for the measuring rod — the standard a thing is held against.

AI agents are in their loudest phase: everything claims full autonomy, and the real is tangled with the marketed. You cannot tell them apart from the landing page. So we measure. Every agent here is held against real signals — who actually uses it, which number is moving, what it charges, which job it genuinely takes off your plate. We strip out the hype and keep what holds up: the canon of AI agents.

Spelled with a K, not a C — we are not the camera company. We measure agents.

What counts as an "AI agent" here

We use a wide definition: anything that autonomously completes multi-step tasks for you — coding agents, workflow automation, research agents, support and sales agents, voice and browser agents, and vertical tools built on top of models.

We do not filter by who shipped it. A frontier lab's release and a two-person team's side project are judged the same way: is it rising?

Currently 19,964 AI agents are live in the index, 19,662 of them with a "what job does it do" teardown.

In the last 24 hours, 244 new products entered the observation pool.

Where the data comes from

Product Hunt · Hacker News · GitHub · HuggingFace Spaces · Reddit · Lobsters · TrustMRR · plus open community submissions. Each source covers a different stage: Product Hunt for launch-day heat, Hacker News and Lobsters for technical credibility, GitHub and HuggingFace for developers voting with their feet, TrustMRR for real revenue.

Watching only one source means you're always late or blind to half the market.

What we break down for each agent

If we can't read a field, we leave it empty. We don't guess and we don't let a model invent a plausible-looking number. A fabricated price is worse than no price.

How to use it