Agents handle repetitive CRM work automatically - updating deal stages after calls, routing new leads, posting weekly pipeline summaries - without manual input. Configure an agent once and it runs in the background every time its trigger fires.
Agents are available to all Clarify users. Limits on the number of agents per workspace may be introduced as the feature matures. To get started, click the Agents link in the left sidebar.
New workspaces are offered four ready-made agents during onboarding - Inbox Reply Drafter, End of Day Wrap-Up, Post-Meeting Follow-Up, and Post-Meeting CRM Update - which you can turn on in one click. You'll find these and more in the agent template gallery at any time (see What you can build below).
Connect the tools a template needs before you turn it on. The email-based templates need an email account connected under Connectors. End of Day Wrap-Up runs on a schedule whether or not one is connected, so without email it fails at the send step every weekday and uses credits on each attempt.
What is an agent?
An agent is a background worker in Clarify built around three things:
Trigger - what causes the agent to run (a record change, a time schedule, a webhook, or another CRM event)
Instructions - what the agent does each time it runs
Tools and access - which data and integrations the agent can use
Unlike Rep, which responds to your messages, agents run automatically without any input from you.
How agents work
Agents run automatically in the background. Once configured, they fire whenever the trigger condition is met.
Go to Agents → select Add Agent - describe what you want in plain English (for example, "after every sales call, update the deal stage and log next steps"), or choose a template from the gallery
Rep builds the agent - it pre-fills the trigger, writes the instructions, and selects the tools based on what you described
Review it - agents Rep builds for you are enabled on creation and start running when their trigger next fires. Review the trigger, instructions, and tools, and toggle it off if you'd rather hold off
It runs automatically - every time the trigger fires, the agent executes its instructions using the tools you've allowed
Stay informed - receive in-app and Slack notifications when the agent acts or needs your approval. Failed runs and approval requests are emailed to you too
Available triggers
Add filters to any trigger to narrow when the agent runs - for example, trigger only when a deal moves to a specific stage, or only for records assigned to you.
When you edit a trigger, Clarify shows how often it is expected to fire and the estimated weekly credit cost, so you can see the volume before saving. The same estimate also appears at the agent level.
Trigger | Description |
Record created | Fires when a new person, company, deal, meeting, or custom object is created |
Record updated | Fires when a record's fields change |
Record deleted | Fires when a record is deleted |
Added to list | Fires when a record is added to a specific list |
Removed from list | Fires when a record is removed from a list |
Relationship set | Fires when a relationship is created between two records |
Relationship changed | Fires when an existing relationship between records changes |
Webhook | Fires when an external service sends data to a unique URL generated for your agent |
Schedule | Fires on a recurring schedule (e.g. every weekday at 9am) |
Schedule is not yet offered in the trigger picker for most workspaces, so you will not see it listed when you add a trigger yourself. Agents installed from a template can still run on a schedule - End of Day Wrap-Up runs every weekday evening - and they do so whether or not the picker offers it.
A trigger that watches meetings needs calendar read access on the connected account behind the agent, and a trigger that watches messages needs email read. If that access is missing, the trigger will not fire. Reconnect the account and grant it - see Accounts.
Get your webhook URL
To trigger an agent from an external service, add a webhook trigger and copy the URL Clarify generates for it.
Open the agent and go to Triggers.
Click Add triggers and choose On inbound webhook, under Other.
Save the agent. The URL is generated only after the agent is saved.
Copy the URL shown under “We’ve generated a custom webhook URL for you to send requests to” using the copy button beside it.
Send (POST) requests to this URL from your external tool to start a run. The data in each request is available to the agent in its instructions.
Create an agent
Select Add Agent from the Agents screen.
Describe what you want - type what the agent should do in plain English, or choose a template from the gallery to start from a pre-built agent.
Answer up to three clarifying questions - if your description leaves details open, Rep asks follow-up questions with suggested options: for example, what counts as a high-value deal, how you want to be alerted, and whether the agent should act automatically or ask for your approval first. Select an option or type a custom answer.
Wait while the agent builds - Rep checks for existing agents, fetches your CRM schema, and generates the trigger, instructions, and tools configuration. The finished agent appears on the right side of the screen with an auto-generated name and description.
Review - agents Rep builds for you are enabled on creation. Adjust the trigger, instructions, or tools if anything doesn't look right, or toggle Enable off to pause it before it runs. You can also enable or disable any agent from the card's overflow (⋯) menu.
Human-in-the-loop
Set any tool or action to require your approval before it runs. The agent will pause, notify you, and wait. Approve or deny in-app - no need to re-run the whole agent.
When an agent needs your input, you'll be notified three ways:
In-app notification - a prompt appears describing the specific action the agent wants to take - for example, "Update campaign 'Demo Outreach Campaign'?" Choose Allow once, Always allow (skips the prompt for that action going forward), or Deny
Slack - notifications can go to a Slack DM or post directly to a channel. From a Slack DM, you can approve or decline a pending action directly, without opening Clarify. If the agent asks you a question instead, the DM carries an Answer button that opens a form covering every question at once, and the run continues once you submit. Only the person who started the run can answer. Configure this in Connectors → Slack.
Sound - an audio chime plays when the agent needs your attention.
Tools and access
Agents are scoped to their creator's permissions. An agent can only see records, emails, and integrations that its creator has access to.
Choose a model
A complex agent needs deeper reasoning. A simple, high-volume agent doesn't, and pays for capability it never uses. Set the model per agent to match the work.
Open the agent and scroll to Advanced, at the bottom of its configuration page.
Click the Model dropdown.
Choose Smart or Fast.
Smart - most capable. Best for complex reasoning.
Fast - quicker and cheaper. Best for simple, high-volume tasks.
If you don't set one, the agent runs on Smart.
Only the agent's creator can change its model. Everyone else sees the control disabled, including workspace admins.
Run agent
To test a new agent or kick off a run outside the normal trigger schedule, use the Run agent button inside the agent. Each manual run counts against your credit usage and rate limit.
There's no live stream of a run. Runs happen in the background and usually take a few minutes - longer for research-heavy agents. You can open a run in the agent's activity log while it's still going, but that view doesn't refresh on its own; switch tabs and back, or reload, to see how far it has got. You'll get a notification when the run finishes or needs your approval, and results appear on the records the agent touched.
If a run is taking longer than usual, don't start a second one. A re-run charges credits again and can repeat actions the first run already took.
Activity log
Every agent has an activity log showing a full history of its runs - when it ran, what triggered it, and what actions it took. Expand any run entry to see the agent's step-by-step reasoning. Credit cost is shown as an average across the agent's recent runs, not for each run individually.
Limitations
⚠️ There is no undo for agent actions. If an agent updates a record incorrectly, open Rep in the agent's chat and ask it to revert the specific change. This is the most significant limitation today. Use approval mode until you're confident in an agent's behavior - it lets you review each action before it executes.
Additional current limitations:
No cancel mid-run. Once a run starts, it runs to completion. Enable approval mode if you're concerned about an agent acting too broadly.
Historical data is limited. Agents act on new data going forward by default. Processing historical records is possible but capped per run - each run supports up to 30 processing steps, and each step can retrieve up to 25 records. A user-configurable turn limit is coming.
No exportable audit log yet. The in-app activity log is the current record of agent actions.
Rep and agents working together
Rep is the conversational layer on top of everything agents do.
Rep builds agents. Describe what you want and Rep creates it.
Rep debugs agents. If something doesn't work as expected, describe the problem to Rep and it will suggest fixes. When an agent run fails, the failure notification also includes an Investigate with Rep button - clicking it opens a Rep chat with the failed run context preloaded.
Rep answers questions about agent activity. Ask "what did the pipeline review agent do this week?" and Rep will summarize the run history.
Rep can delete agents. Ask Rep to delete an agent by name from chat.
What you can build
Here are examples of agents customers are already running.
Agent | What it does | Who it's for |
Inbox Reply Drafter | When an inbound email arrives, drafts a reply in your voice and saves it to your drafts for review before sending | Anyone who lives in their inbox and wants a head start on every reply |
Post-Meeting CRM Update | After every meeting, reads the transcript and updates deal stage, next steps, and custom fields automatically | Sales reps who spend time on manual CRM updates after calls |
Post-Meeting Follow-Up | After a customer call, drafts a follow-up email and saves it to your email drafts for review before sending | Account executives and CSMs who send follow-up emails after every call |
End of Day Wrap-Up | Every weekday evening, recaps what moved in your pipeline, what you touched, and what needs attention tomorrow | Founders and GTM leads who want to close out the day with a clear picture |
Weekly Pipeline Digest | Every Friday afternoon, posts a structured pipeline summary to your Slack channel | Sales managers and founders who want a pulse on the week |
Available tools
Connect tools in Connectors, then set allow, require approval, or deny per tool per agent.
Clarify CRM
The core tool for reading and writing your CRM data.
What an agent can do:
Read company, person, deal, meeting, and task records
Update fields, deal stages, labels, and custom fields
Create new records (companies, people, deals, tasks)
Deduplicate and merge records
Route new leads to the right owner based on geography, size, or segment
Log structured notes and follow-up tasks from meeting transcripts
What it cannot do:
Access records the agent owner doesn't have permission to view
Delete records - unless you explicitly add the
delete-recordstool, which makes deletion permanentModify workspace settings or user permissions
Add someone to a campaign directly
Getting people into a campaign: an agent can't enroll someone into a campaign directly, and asking it to will not work. Campaigns take their recipients from a list. So the pattern that does work is: point the campaign at a list, then have the agent add people to that list - the campaign enrolls them as they arrive. See Campaigns.
Email (Gmail or Outlook)
Connects to your email account for reading and drafting email.
What an agent can do:
Read inbound emails and identify opportunities or required responses
Draft email replies and save them to your email drafts for review
Send emails automatically when configured to do so
Create deal records from email buying signals
Draft outreach or follow-up emails based on CRM data and web research
What it cannot do:
Draft or send emails from a colleague's inbox - only from the agent owner's account and connected email aliases
Access emails in other connected accounts
Delete emails
Send formatted HTML or tables - agent emails are plain text
Formatting: agent-written emails can include bold text, links, bullet lists and numbered lists, and a section heading arrives in bold. What they cannot include is a table - there is no table support at all, so anything the agent builds as one arrives as raw tags, and the same goes for any other HTML it writes. Asking the agent in its instructions to "use a table" or "don't use HTML" does not change this. If you need a formatted layout, have the agent draft the email and format it yourself before sending.
Slack
Connects to your Slack workspace for posting messages and reading channel activity. To connect Slack, go to Connectors and click "Connect" on the Slack tile.
What an agent can do:
Post messages to any Slack channel the owner has access to
Send a direct message to yourself or others
Post structured summaries, briefings, and pipeline reviews
Read Slack messages and react to events in Slack
What it cannot do:
Post to channels the agent owner isn't a member of
Send messages as another user
Create or archive channels
Granola
Connects to your Granola meeting notes for reading transcripts and extracted content.
What an agent can do:
Read meeting notes and transcripts created in Granola
Extract action items, decisions, and key topics from notes
Use meeting content as context for updating CRM records or drafting follow-ups
What it cannot do:
Create or edit notes inside Granola
Access notes from meetings the owner wasn't part of
Notion
Connects to your Notion workspace for reading and updating pages and databases.
What an agent can do:
Read pages and database entries
Create new pages and database records
Update existing page content
Write documentation, briefings, or summaries directly into Notion
Sync CRM data into a Notion database
What it cannot do:
Access pages the agent owner doesn't have permission to view
Delete pages or databases
Modify workspace settings or sharing permissions
Linear
Connects to your Linear workspace for reading and creating issues.
What an agent can do:
Read existing issues, projects, and team data
Create new issues from meeting notes, emails, or CRM activity
Add comments to issues
Update issue status and assignees
What it cannot do:
Delete issues
Modify team or workspace settings
Access issues in teams the agent owner doesn't belong to
Web search
Built-in - no setup required.
What an agent can do:
Search for company news, funding announcements, and job changes
Research a prospect or account before outreach
Pull recent information to add to a briefing or summary
What it cannot do:
Access pages behind a login or paywall
Submit forms or interact with websites
Store or remember search results across runs
Custom MCP servers
For teams that want to connect agents to tools not listed above.
What an agent can do:
Connect to any remote MCP server your workspace configures (OAuth or API token over HTTP)
Use any tools exposed by that server within agent instructions
What it cannot do:
Connect to local npm/stdio MCP servers (not yet supported)
Bypass the tool permission settings configured in Connectors
You don't need to contact support to set this up. A workspace admin adds the server once under Connectors, and each person then authorizes their own account on it. An agent uses the connection belonging to whoever owns it, so connect it as the person who owns the agent you're building. See Rep Connectors for the setup steps and the three authorization options.
Two things to know once the connector exists. Adding it to an agent doesn't give the agent any of its tools yet - open the agent's Tools section and add the specific tools you want it to use. And each tool that writes or changes data arrives set to Ask permission, so an agent running on a schedule or a webhook will pause and wait for you the first time it reaches that tool. Set those tools to Always allow if the agent is meant to run unattended.
Sharing and visibility
All agents are visible to everyone in your workspace. Anyone can see the agent's name, trigger, instructions, and tools.
The private/public toggle controls who can see the agent's activity - its run history, reasoning steps, and Clarify chat. Set it to public to let teammates follow along; keep it private if the agent's activity contains sensitive information.
Only the creator can edit an agent's instructions, triggers, tools, or toggle it on or off.
Non-owners can see whether an agent is enabled or disabled, shown as a read-only status. Controls for resuming a paused run are owner-only; everyone else sees an explanatory tooltip.
Workspace admins have the same access as any other user. They can view all agents and their activity but cannot edit agents they don't own.
Anyone in the workspace can duplicate an agent. The person who duplicates it becomes the new owner - it runs using their permissions and tool connections.
To delete an agent, open it and choose Delete from the ⋯ menu beside its name. Only the agent's creator sees this option. You can also ask Rep: "delete [agent name]."
Credits and billing
Agent runs consume credits from your workspace credit pool. Costs are based on actual AI usage per run - see Example agent costs below for typical ranges. Credits draw from the same pool as the rest of Clarify. For included credits, on-demand pricing, and your current per-credit rate, see Billing, pricing and payments and your in-app billing page.
A run that can't finish still costs credits. It is charged for the work it did before it stopped, so an agent that fails every time it runs keeps drawing your balance down. If an agent reports a failed run more than once, open its activity log and fix the cause or turn the agent off.
How agent credit costs work
Agent costs are based on actual AI usage per run. Shorter, simpler interactions cost less than longer, multi-step ones.
Factors that affect cost:
Amount of data read - querying a few records costs less than scanning your entire pipeline
Number of actions taken - a single comment costs less than creating records + sending emails + posting to Slack
Conversation complexity - multi-step reasoning with back-and-forth tool calls uses more than a single lookup
Example agent costs
Agent type | Typical cost per run | Runs per 1,000 credits |
Inbox Reply Drafter | ~20 credits | ~50 |
Post-Meeting CRM Update | ~50 credits | ~20 |
Post-Meeting Follow-Up | ~50 credits | ~20 |
End of Day Wrap-Up | ~60 credits | ~16 |
Weekly Pipeline Digest | ~100 credits | ~10 |
Costs are estimates. Actual usage varies based on data volume and conversation length.
Research-heavy agents cost more. Agents that use web search, scan large amounts of data, or process a backlog on their first run can run far beyond these ranges - hundreds or thousands of credits in a single run. The table covers typical light agents, not research- or volume-heavy ones.
Estimating your monthly cost
Clarify estimates this for you - open the agent, or edit one of its triggers, to see the expected firing frequency and weekly credit cost. To sanity-check it yourself, multiply credits-per-run by how often the agent fires. The Weekly Pipeline Digest runs ~4×/month (~400 credits). A Post-Meeting agent firing after ~3 meetings/day costs roughly 3,000 credits/month. The Inbox Reply Drafter scales with email volume - it’s the highest-usage agent of the five.
Tracking usage
Monitor agent credit usage on your billing page. The credit usage breakdown lists each agent as its own row, so you can see which one is spending the most.
If a run reaches your workspace credit limit mid-run, it pauses instead of failing. Look for the resume banner on the agent and continue the run once credits are available.
Keep agent costs down
A few habits keep a high-volume agent from draining your credits:
Check whether Clarify already does it. Some work is already automatic and free - tasks created from a recorded meeting link themselves to the deal and person on that meeting, with no agent involved. An agent repeating built-in work bills on actual AI usage every run. See Tasks.
Match the schedule to the need. An agent set to run hourly costs far more than the same agent run once a day. Widen the schedule, or add trigger filters so the agent only runs when it has to.
Check the per-run cost early. Agents Rep builds are enabled on creation, so review a new agent's first run in the activity log - especially one that uses web search - to confirm what a single run costs before leaving it on its schedule.
Keep on-demand credits off, or cap them. On-demand credits are off by default, so credit features pause when your monthly credits run out instead of charging you. If you turn them on, set an On-demand spend limit. Email alerts arrive at 80%, 90%, and 100% of your monthly allowance. See Billing, pricing and payments.
Run new agents in approval mode. Require approval while you learn how an agent behaves, so it pauses before each action instead of running uncapped.
Match the model to the work. A high-volume agent doing simple work runs cheaper on Fast. See Choose a model above.
Rate limits
Each agent is limited to 100 runs per hour. This caps how often an agent fires, but not how much a single run costs - a research-heavy run can still be expensive, so use the cost controls above. To further limit a single agent, set a lower run cap (runs per hour) at the bottom of its configuration page.
FAQ
Can I stop an agent once it's running?
Not yet - once an agent run starts, it runs to completion. Canceling a mid-run agent is on our roadmap. In the meantime, enable approval mode so the agent pauses and asks before taking any action.
What happens if an agent updates a record incorrectly?
There's no undo button in the notification or activity log. To undo a change, open Rep in the agent's chat and ask it to undo the specific change. This is the most important limitation today - see the Limitations section above for more detail.
Who can see my agents?
All agents are visible to everyone in your workspace. The private/public toggle controls who can see the agent's activity (run history and chat logs) - not the agent itself. Only the creator can edit or toggle an agent.
Can my teammates use my agents?
Anyone in your workspace can view and duplicate your agent. When someone duplicates it, they become the owner and it runs using their permissions and tool connections. Agents always run as their owner - access to records and integrations is scoped to whoever owns the agent.
Will I get notified when an agent runs?
Only when an agent acts or needs your approval. You'll receive an in-app notification, plus a Slack notification (DM or channel, based on your settings) if your Slack workspace is connected. Two of these are emailed to you as well: when a run fails, and when a run is waiting on your approval. To stop those emails, turn off the Email toggle for Agent run failed or Agent awaiting input in Settings → Account → Notifications. View the full run history and records touched in the agent's activity log.
Can agents send emails on my behalf?
Agents can draft emails and save them to your email drafts for review - or send automatically if configured to do so. All sends go via your email account. An agent cannot draft or send from a colleague's inbox.
What integrations do agents support?
Agents connect to Slack, Notion, Linear, Gmail and Outlook, Granola, and any custom MCP server your workspace configures. Web search is also available as a built-in tool. Connect tools in Connectors, where you can set allow, require approval, or deny per tool per agent. More connectors are on the roadmap.
Can agents process my historical CRM data?
Agents act on new data going forward by default. Processing historical records is possible but limited by a per-run cap. Contact support if you need to process a large historical backfill.
How do credits work?
Agent runs and Rep usage both draw from your workspace credit pool. Costs are based on actual AI usage - shorter, simpler runs cost less. See the Example agent costs table above for typical ranges. Your billing page breaks credit usage down per agent; the cost of an individual run is not shown.
Can an agent trigger another agent?
Yes - if an agent creates or updates a record, that change can trigger another agent. Clarify automatically prevents recursive loops: after five chained triggers, the chain stops and no further agents are triggered.
What happens if my agent hits the rate limit?
Each agent is limited to 100 runs per hour. New trigger events are dropped until the hourly window resets. The agent resumes on the next trigger after the reset.
What plan do I need to access Agents?
Agents are available to all Clarify users on any plan. Limits on the number of agents per workspace may be introduced as the feature matures.
Where do I find Agents in Clarify?
Agents are accessible via the Agents link in the left sidebar.