Agent is the main tab where you work with our global agent. Tell the global agent what you’re trying to do, and it works across our products to help you get it done.
You can also open the global agent from the Agent icon in the top right bar, which opens the chat interface within the page you’re working on. Agent is available in Workflows, Search, or the home screen.
Global agent capabilities
Our global agent allows you to easily interact with Clay’s existing products.
Global agent vs Sculptor
Previous Sculptor entry points in Workflows, Search, and the home screen now open the global agent. Sculptor has not moved in tables or in Claygent Builder, where it keeps its own chat history.
So in a table and in Claygent Builder you may see two chats. They differ in what each one is for:
- The global agent analyzes a table, and rebuilds one as a workflow — hand it several tables and it will consolidate them into one. It cannot build or edit a table for you.
- Sculptor builds and edits a table, and writes a Claygent prompt.
See Sculptor for what it does inside a table and in Claygent Builder.
Global agent vs Clay CLI
Our global agent runs on the same engine as the Clay CLI, so anything you can do from the command line you can do here, with Clay’s own interface around it. See Clay API and CLI for the command-line route. Which one you reach for comes down to how you want to work.
For the tools themselves, see Workflows, Audiences, Search for Find leads, Sequencer, and Signals. Account Agents in Workflows covers the workflow step that researches one account at a time, and when to add one.
Starting a task
- Open
Agentin the left nav, then selectNew task. The start screen asksWhat should we work on today?.- The suggested prompts are based on your workspace context.
- Say what you want to end up with, not just what you want to look at. “Save these to a segment” and “build this as a workflow” get you something you can use; “show me some accounts” stops at a list.
- The global agent’s own examples are shaped this way:
Route inbound leads,Build a prospecting workflow,Score and prioritize accounts,Build and enrich a lead list.
- The global agent’s own examples are shaped this way:
- Say what to leave out. Accounts you already own, and anyone you’ve already engaged — naming them up front is what keeps them out of the results.
- Answer the questions it comes back with. The global agent asks for the detail you left out before it starts building, and you can
Skipany question you’d rather not answer. Closing the question card instead stops the task. - Open what it built. The
Artifactspanel collects what the global agent made in this task.
Scheduling a task to repeat
You can ask the global agent to make any task recurring, so the same work runs on a cadence instead of once — daily, on weekdays, weekly, or monthly. Recurring tasks live under Scheduled in the Agent sidebar.
Each schedule keeps the Instructions it follows, its Next run, and every run it has had, so you can read back what changed between one run and the next. A run that needs an answer from you waits rather than guessing, and the schedule stands still until you reply.
You can pause a schedule, run it early, or end it from the schedule’s own page.
Giving the global agent more context
The global agent starts from your Knowledge Hub, your workspace’s record of what you sell and who you sell it to. When a task should lean on something specific, add a reference to it as you write the message:
- Type
@for a part of yourKnowledge Hub, or a document you’ve uploaded there. - Type
$or/for yourWorkspace skills, the instructions your team has written for work the global agent does often. Your team creates them on theAgent skillspage in workspace settings. - Select
Attach fileto add one or more files to a message: spreadsheets including CSVs, PDFs, Word documents, text files, and images, up to 50 MB each.
Note: One message can carry up to 50 references. Add more and the send button stays locked until you remove some.
See Knowledge Hub for what lives there and how to change it.
Permissions and approvals
Every task and every schedule carries a permission level that decides what the global agent does without stopping to ask. Ask for approval is the default. Set it in the message box before you send a task, and on a schedule’s own page for runs nobody is watching.
Note: a task or schedule set up before View only was retired still shows it. Nothing breaks, and saving a change to that task or schedule means picking one of the two levels above.
What waits for your approval
Before you leave a task running, it’s worth knowing which actions stop for your approval under Ask for approval:
- Running a workflow, test data included, or a single step of it as a test.
- Running a workflow on a trigger’s existing records: a CSV’s rows, a
Find leadssearch’s results, or a segment’s members. - Re-running a run, or resuming one that’s paused.
- Publishing a workflow, which turns on its live triggers so runs start from new events, or turning a paused trigger back on.
- Running a function or an enrichment on its own, outside a workflow.
- Creating, changing, pausing, resuming, or deleting a signal.
- Archiving a segment, deleting a segment field, or deleting a workflow, a workflow alert, a Claygent, or a schedule.
- Accepting or dismissing a suggested change to your
Knowledge Hub.
Plenty of real work, including some that spends credits, still goes ahead on its own: editing a workflow and its triggers, trying out an action or a piece of code before it goes into a workflow, searches, and creating or updating segments and fields. So do deleting a webhook and taking a destination off an ad sync.
The same goes for renaming an API key and starting a credit purchase, which you still complete yourself. Sending a test email from a campaign, or a test event to a webhook, needs no approval either.
What the global agent won’t do on its own
The global agent works through a task without stopping for approval at each step, so it’s worth knowing where it stops on purpose. The actions that reach people outside your workspace, or that you can’t take back, stay with you:
- It won’t launch or send a campaign. The global agent drafts one, edits the copy and settings, runs a spam check, and sends a single test email. Starting it, pausing it, and completing it happen in Sequencer, by you.
- A signal it creates starts paused, so you can read it back before it starts checking records and spending credits. It resumes the signal once you confirm, not on its own initiative.
- It won’t create or edit a table. Ask it to rebuild one as a workflow instead.
- It won’t delete a campaign or an ad sync. It can take a destination off an ad sync that’s paused or still a draft, or off an always-on sync that failed.
Under Ask for approval, a workflow, function, or enrichment run waits for your approval, so check the global agent’s work before you select Allow once on the approval card: open the Artifacts panel and look over the workflow or segment it produced.
It does come back to you mid-task when a decision is genuinely yours and would change the outcome — which list to start from, which of two scoring approaches to take. It gathers those onto one card rather than interrupting you repeatedly.
Credits and pricing
The global agent spends data credits as it works, and the work it sets up spends credits and actions the same way it does when you run it yourself. Two kinds of spend sit behind a task:
- The agent’s own work. Reading your workspace, answering a question, and planning an approach cost very little. Building something — a workflow, a campaign, a segment — costs more, because it takes more steps across more of Clay.
- What the agent runs for you. Enrichments, workflow runs, and other downstream work spend credits and actions as usual. That spend lands under whatever the agent built, not under the agent.
Working with the global agent spends data credits only; it doesn’t consume actions. What you pay for its own work doesn’t change with the model behind it, so a model change on our side never changes your cost.
Estimated credits for common tasks
These are rough estimates of what the global agent’s own work spends on some common tasks. Your costs will vary with what a task involves, how long the conversation runs, and how much data the global agent works through.
The estimates don’t include enrichments or other work the global agent runs for you, which spend credits and actions as usual.
Bonus credits in Beta
Every eligible workspace gets a one-time grant of bonus credits when Beta starts. The grant lands the first time anyone in your workspace starts a task with the global agent.
Bonus credits are added to your workspace’s data credit balance and spend like any other credit — on the global agent or anywhere else in Clay. A workspace on a trial gets no grant, so the global agent spends the credits the workspace already has.
Keeping an eye on spend
When the global agent stops for your approval, the card carries an Estimated cost in credits and action executions before anything runs. Some actions can’t be priced ahead of time, and the card says Cost can’t be estimated when that happens.
The permission level you set decides how often you see a card at all, so it’s also the lever on how much runs unwatched. A schedule spends credits on every run, which is worth weighing when you set its level.
To see what the global agent has spent, select your account name, go to Settings then Usage, and open the Agent tab. It breaks spend down by task over the period you pick. The work the agent set up shows up in the other tabs, under the workflow or table it belongs to.
If your workspace runs out of credits, the global agent stops: you can’t start a new task or send a follow-up until the balance is topped up. Clay tells you so in the task, with Add credits alongside.
See Actions & Data Credits for how the two meters work, and Credit usage for the rest of the dashboard.
FAQs
Why is this taking a few minutes?
Work that spans several steps runs for as long as it needs and shows its progress as it goes, so a task that builds something takes longer than one that answers a question. Your task list marks it Working while it runs, and Unread when there’s something new.
You don’t have to wait it out — send your next message while it runs and the global agent queues it, marked Sends after this run. Stop ends a run you no longer want, and you can send again straight away. If a run fails, the partial output is saved and the task offers Retry.
Can my teammates see a task I started?
Only if you share it. Select Share task, turn on Share with workspace, and copy the Task link.
Anyone in the workspace with that link can read the task’s saved messages. They can’t send anything in it, and only the person who started a task can rename it.
Does the global agent know which page I’m looking at?
Yes. The global agent can see the page you have open, so you can ask about what’s in front of you — “score the accounts in this segment”, “why is this workflow failing” — without naming or linking it first. Opening the global agent inside a workflow, an audience, or a search also changes the suggested prompts to match.
Can I speak a message instead of typing it?
Yes. Select Dictate, the microphone in the message box, and speak. It works when you start a task, when you follow up, and when you answer one of the global agent’s questions.
When you finish, what you said lands in the message box as text you can edit, and nothing sends until you send it. A recording finishes on its own after five minutes.
Why don’t I see Agent in my workspace?
The global agent is rolling out gradually during Beta, so you may not see it yet even once it has reached other people.
If you’re an Enterprise customer and want to try it, ask your Growth Strategist for access. On a Growth or Launch plan, sign up for the early adopter program, then have a Workspace Admin turn on Beta program in your workspace settings. Clay notifies you in the product once you have access.
Explore other docs
Other popular resources
Join a cohort, learn Clay fast!
The faster way to master Clay. Sign in if you're enrolled in a cohort (current or past) or apply!
Hire GTME Talent
Find and connect with GTM talent who've demonstrated expertise in building advanced workflows




