Claygent, explained

Claygent is the AI research agent built into Clay. You add it as a column in a Clay table, write a prompt, point it at input columns from the row, and define the fields you want back. It then searches the web, opens pages, reads them, and writes a structured output into the row. One row equals one run.

Clay describes Claygents as agents that take inputs, follow your instructions, and write a structured output. In practice teams use them for four jobs: account research and meeting prep, finding data no provider sells as a field, qualifying a list against custom criteria, and generating personalized outbound copy grounded in what the agent found.

It is available on every plan tier. Account Agents, the always-on variant that watches an account list rather than running once per row, is in beta and limited to Launch, Growth, and Enterprise.

Claygent lets you pick the model per column, and the choice is the single biggest lever on your bill. Clay's own managed models are priced at a flat credit rate per row: Helium at 1 credit, Neon at 2 credits, Argon at 3 credits. Argon is the deepest of the three and the one people default to, which is exactly how a Clay bill triples without the workload changing.

Beyond the managed models, you can swap in frontier models such as Sonnet 5 or GPT 5.6, or open-weight models such as Kimi K2.6 and GLM-5.2. Those run at variable pricing rather than a fixed credit rate, so the same column can cost wildly different amounts depending on how much the model reads before it answers.

The sane default: Helium for simple lookups where the answer is a fact on a homepage, Neon for qualification calls that need a bit of judgment, and Argon only for genuine multi-step research. Set the model per column, not once for the whole table.

Until March 2026 a Claygent run only spent Data Credits. Since then, every run spends a Data Credit and an Action simultaneously, so one research task draws down two separate balances. Anyone budgeting off the old model is under-counting.

Concretely, on the Launch plan at $185 a month you get 2,500 Data Credits and 15,000 Actions. Research at Argon depth costs 3 credits per row, so the plan caps out near 833 researched companies a month: roughly $0.22 per company. Actions are not your constraint; Data Credits are. Growth at $495 a month moves the ceiling but not the shape of the problem.

Two things people underestimate. Failed and empty runs still burn credits, so a badly tuned prompt costs money while it teaches you nothing. And Claygent has a real learning curve: prompt tuning per use case is normal, and most teams running Clay seriously have a dedicated operator whose salary dwarfs the subscription.

Claygent is a search-and-read agent. It handles the open, indexable web well. It stalls on sites that require driving: JS-heavy careers pages that render nothing to a fetch, filtered directories where the data appears only after you set the filters, gated pricing, portals behind a login. On those, the agent tends to come back with a confident summary of the homepage instead of the answer you asked for.

It is also structurally row-based. Claygent enriches rows you already have. It does not discover candidates you have never heard of, dedupe them, and qualify them in one motion. Building a list from scratch in Clay means wiring several sources and columns yourself.

Finally, the proof is thin. You get source links in cells. If your process requires showing a stakeholder why a company was scored the way it was, links to pages that may have changed since are not the same thing as dated evidence.

Stay on Claygent if you already live in Clay, have an operator, and want research to sit next to your waterfalls, routing, and CRM sync in one table. That integration is real value and no API replaces it.

Switch to a research API when the research is the whole job and you want it programmatic: a plain-English brief in, a typed JSON verdict out, with citations attached and one fixed price per company. Gumshoe does exactly that, drives a real browser when a site needs clicking rather than reading, and bills one credit per company: $99 a month for 2,000 cases, versus roughly 833 Argon rows for $185 on Clay's Launch plan. One meter instead of two, and no operator required.

The decision is not "is Claygent good". It is whether you want a workbench you operate or an endpoint you call. If you bought Clay for one AI column, you are paying for a workbench you barely use.