Skip to main content

​
What is an Agent

An Agent is an autonomous AI worker. Instead of a single Prompt, you give it a purpose, the inputs it receives, the outputs you expect, and the tools it can use. The Agent then reasons, uses those tools and completes the task on its own. Use an Agent when you can’t lay out the steps in advance: the path depends on what the Agent finds along the way, it needs to call external services, or it works through a back-and-forth conversation. For a single well-defined instruction, a Canvas is simpler. If you can draw the steps in advance, a Workflow is more predictable and easier to test, and it can still use an Agent for one of its steps.
Agent creation is enabled per Workspace. If you don’t see the Agent option on your Dashboard, contact your Tela team.

​
Creating an Agent

Unlike Canvases and Workflows, Agents are created directly inside a Project, so they don’t go through the Draft step. From your Dashboard:
  1. Click Agent
  2. In the New agent dialog, pick the Project the Agent belongs to
  3. Choose the harness, the engine that runs the Agent: Lightweight · Tela is faster and works with models from any provider, and Claude Code · Anthropic is more thorough for complex work, with Claude models. The harness can’t be changed after the Agent is created
  4. Click Continue, then pick a template to start from or click Start from scratch
New agent dialog with the project and harness options
You can also create an Agent from a Project’s page through the Create button.

​
Understanding the Agent page

Like a Canvas, an Agent has Craft, Test and Usage experiences in the top center. The Craft experience is organized in sections on the left sidebar:
  • Purpose: the instructions that describe what the Agent does and how it should behave. This is also where you pick the model
  • Inputs: the variables the Agent receives on each run, as text or files
  • Outputs: the structured output you expect back, if any
  • Skills: specialized capabilities you add for complex or recurring tasks
  • Subagents: helper Agents that your Agent can delegate parts of the work to
  • Files: reference documents the Agent can consult while working
  • Integrations: external services the Agent can call without ever seeing their credentials
Agent page in the Craft experience
Start with Purpose and Inputs only. Add Skills, Subagents and Integrations once the basic run works the way you expect.

​
Running your Agent

Use the Test your agent panel on the right. Fill in the Variables (the Agent’s Inputs) and send a message in the box at the bottom. The Agent works through the task and answers in the same panel, and you can keep the conversation going: it keeps the context between messages.

​
Testing and publishing

When a run looks good, click Review to save it as a test case, or go to the Test experience and click Create first test. Each test has inputs, an expected output and optional instructions for the evaluator that judges the result. When you are ready, click Publish. Publishing creates a stable version that your integrations use. Your team can also use the published Agent through an App. To call the Agent from your own systems, open the … menu and choose Connect via API.