Getting Started
Deploy Your First Outrun Agent
Start with one narrow customer outcome. This guide takes you from choosing a specialist role to reviewing its first activity. Exact connector options depend on the tools available in your workspace.
1. Choose the outcome
Pick a job with a clear input, decision, and next action. Good starting points include:
- answering product questions with a Pre-Sales Agent
- applying your criteria with a Lead Qualification Agent
- moving a qualified opportunity forward with a Sales Closer
- reviewing accounts and follow-up work with a Customer Success Agent
Start with one role rather than an end-to-end process. You can add handoffs and agent groups after the first workflow behaves the way your team expects.
2. Create your workspace
Visit app.getoutrun.com, sign in, and create or select the workspace where this agent should operate. A workspace separates its members, agents, connections, and activity from other organizations.
3. Connect the required context
List the minimum information the role needs: for example, product knowledge, account records, a shared inbox, or deal data. Then connect the supported source or tool for that information.
Follow Connect sources for the general setup flow and use the relevant integration page to confirm whether that connector can retrieve data, run actions, synchronize records, or provide more than one of those capabilities.
Connecting a tool does not imply that every read, write, or sync operation is available. Confirm the capability required by your workflow before deployment.
4. Create the agent
Open the AI area in Outrun and create a new agent. Choose the template closest to the job you selected, then configure:
- the goal and boundaries of the role
- the knowledge and connected tools it can use
- any play or workflow instructions it should follow
- the situations where it must ask for help
Use concrete instructions and examples from your real process. Avoid giving the first version more tools or responsibilities than it needs.
5. Keep human approval on
Require review for customer-facing messages, external writes, or other sensitive actions while you validate the workflow. Human-in-the-loop controls let a person inspect the proposed action before it is completed.
See Human in the loop for approval options and operating guidance.
6. Test a representative scenario
Run a case your team understands well. Check that the agent:
- found the right customer and business context;
- followed the role's instructions and boundaries;
- selected an appropriate next action;
- requested approval where expected; and
- produced an activity trail your team can understand.
Test an edge case too, especially a situation where the agent should stop or hand work to a person.
7. Review and correct
Inspect the activity and proposed actions rather than judging only the final response. When something is wrong, correct the instruction, connected context, or approval rule responsible for the result. Re-run the scenario to confirm the change.
8. Deploy, observe, then expand
Put the agent into a limited real workflow and review its early activity. Once the role is reliable, you can:
- add another specialist and coordinate them with an agent group;
- automate recurring work with loops;
- add a customer-facing surface with the chat widget; or
- connect more context and actions as the workflow requires them.
Next Steps
Need help? Email [email protected] with the agent name, the activity you reviewed, and what you expected to happen.
