Managed Agents From a Prompt
Describe the job, connect the tools it needs, and set its permissions. Managed Agents can remember useful context, split complex work between specialists, and follow up on a schedule. For research and file work, enable Computer and run the task in chat.
- Build from a prompt
- Memory you can review
- Specialists for complex tasks
- Permissions you control
From prompt to published agent
-
Describe the agent
Write what you need in plain English, or start from a use case like Knowledge Base Assistant or Email Campaign.
-
Set the guardrails
Choose events or schedules, approve permissions, attach your knowledge base, and set the capabilities the agent can use.
-
Test before activating
Test routine requests, unclear instructions, and handoffs using your own contact and sample records. Tests can perform real actions. Review the results before publishing, then monitor runs and refine the instructions.
Setup view shown. Workspace layout may vary as the updated interface rolls out.
From one request to repeatable work
Build around a task your team understands. Add memory, connections, or a follow-up plan as the job requires, then inspect what happened in each run.
-
Keep useful context
Memory stores relevant preferences and reusable procedures for recurring work. Inspect what the agent remembers and correct it when your business changes.
-
Give parts of a job to specialists
An agent can assign research, record checks, and other focused tasks to subagents, then review their results. Independent work can run in parallel; dependent steps wait for the information they need.
-
Start from another system
An inbound webhook passes an order, form response, or other event directly to a published agent. Tell it how to interpret the data and which actions it may take.
-
Wait for the right time
Schedule steps around an appointment or add a delay within a run. Conditions and stop rules can prevent follow-ups after a reply, booking, or other change you define.
-
Use current business details
Reference custom values and your business profile in agent instructions. Updated booking links or contact details apply on the next run without republishing the agent.
-
Bring an n8n workflow into Agent Studio
The beta importer prepares an agent from your workflow export. Review the converted steps, reconnect the required apps, and test any logic that changed before activating it.
Research, analyze, and make the file
Give an agent the browser and file tools it needs to gather information and produce a document or spreadsheet. Enable Terminal when the task calls for scripts or calculations.
For chat and manual runs
Computer does not run from workflow or event triggers. Keep scheduled CRM work separate from tasks that need browser interaction.
Review sensitive actions
Configure approval requests for actions such as sending messages or publishing. Payments and deletions always need approval. You can inspect browser activity and take control when needed.
What will you put an agent on?
These are example instructions to adapt and test. The agent needs the relevant connections, knowledge, and permissions before it can do the work.
-
“Prepare a brief before tomorrow's customer call.”
Use approved CRM access to gather context, then schedule a follow-up task after the appointment.
-
“Compare these suppliers and put the findings in a spreadsheet.”
Run it in chat with Browser and Files enabled. Review the sources and the file before using the results.
-
“When an order arrives, find the customer and add an internal note.”
Connect the order system through an inbound webhook and define which CRM records the agent may update.
You decide what it can do
Set access, knowledge, and starting conditions deliberately. Review the results of a test before letting an agent act on customer records.
-
Only the skills you allow
Choose access to contacts, conversations, calendars, invoices, email, and other CRM tools. Review its connected tools and restrict access to the records and actions the job needs.
-
Answers from your own content
Attach approved material as a reference for answers, then test the responses for accuracy. Web search and image generation are separate switches, off unless you turn them on.
-
Triggers you define
Choose a supported event, webhook, or schedule for CRM work. Computer tools have a separate boundary: they work in chat and manual runs only.
-
A person can take over
Write a handoff rule into the instructions so complex conversations go to your team. The agent handles the routine, and your people handle the judgment calls.
Permissions setup view. Grant only the CRM access each agent needs.
See how to scope and test an agent for your work
Bring a task your team handles by hand. We can help you identify the tools, permissions, and review steps it needs.
See what the agent actually did
Inspect a run
Start in Activity, then use Agent Logs to review the conversation, actions, and execution steps. Subagent Activity shows delegated tasks and their results.
Track progress over time
Customize a Metrics dashboard or set a Monitor target for recorded activity. Daily checks and configured email or webhook alerts can flag progress that needs attention.
Agent Health and Self Healing are separate tools for Voice AI and Conversation AI. They prepare prompt improvements for human approval; they do not automatically rewrite Managed Agents. Explore the AI tools.
Managed Agents live in Agent Studio, not AI Studio
Agent Studio
Where you build and run Managed Agents: prompt-based agents that work in chat, from an event, or on a schedule. Growth includes 100 runs per month; Unlimited includes 1,000.
AI Studio
Builds websites and web apps from a description, with a database, user logins, and checkout connected to your CRM. Pay-per-use or covered by AI Employee.
One Managed Agent run is one chat session, one event invocation, or one scheduled invocation. The test area includes 30 free runs per month before production allowances or pay-per-use billing apply.
AI Employee Growth costs $97 per month and includes 100 Managed Agent runs. Unlimited costs $147 per month and includes 1,000 runs; Unlimited is included with Enterprise. Account usage-limit settings control whether applicable pay-per-use service continues after an allowance is used.
Building a site or app, not an agent? That's AI Studio.
Frequently Asked Questions
What are Managed Agents in SMBcrm?
What are Managed Agents in SMBcrm?
Are Managed Agents the same as Agent Studio?
Are Managed Agents the same as Agent Studio?
Do I control what a Managed Agent can do?
Do I control what a Managed Agent can do?
How are Managed Agents billed?
How are Managed Agents billed?
Are Managed Agents the same as AI Studio?
Are Managed Agents the same as AI Studio?
Can a Managed Agent use a browser or create files?
Can a Managed Agent use a browser or create files?
What does an agent remember?
What does an agent remember?
How do subagents work?
How do subagents work?
Can an agent pause and follow up later?
Can an agent pause and follow up later?
Can another app start a Managed Agent?
Can another app start a Managed Agent?
Can I import an n8n workflow?
Can I import an n8n workflow?
Does Agent Health change Managed Agents automatically?
Does Agent Health change Managed Agents automatically?
Ready to put an agent on it?
Start with one task, connect the tools it needs, and review a test run before publishing.
-
60-day money-back guarantee
Eligible Startup and Professional initial subscription fees only. Usage and add-ons are excluded. Read the refund terms.
-
No setup fees
Plan your setup with training and support
-
Monthly or annual billing
Monthly plans renew each month, and you can cancel before the next renewal. Annual plans are paid upfront for the year.