Skip to main content

Creating agents

You can create personalized agents to support day-to-day work: define what they do, connect knowledge and tools, and shape how they respond.

For example, you can build a customer support agent from documentation your teams already use. Once configured, the agent answers common questions consistently.

There are several ways to build the same agent:

  • From chat — In the default chat workspace, describe the agent’s job and refine it through follow-up messages. Devs.ai saves the agent and offers links to open or try it.
  • Simple mode — Describe the job in plain language, then refine through short sections and try the agent as you go. This is the default create experience when it is enabled for your organization.
  • Advanced editor — The full control panel for every setting, including API functions, custom inputs, the full toolbox, and custom MCP servers.

These paths produce the same kind of agent. You can start in chat or Simple mode and open the Advanced editor whenever you need deeper configuration.

📝 Note: Simple mode and the new chat workspace (including create-from-chat) roll out as they are enabled for organizations. If you do not see them yet, create and edit still use the Advanced editor (and classic chat). Contact your Devs.ai representative if you want these experiences for your organization.

Create an agent from chat​

When the new chat workspace is enabled, you can create an agent without leaving default chat:

  1. On the chat home, choose Create an agent, or type a request such as “Create an agent that helps me…”
  2. Describe what the agent should do, how it should respond, and which resources it should use (for example skills, tools, or documents).
  3. When the agent is saved, use Open agent to inspect the configuration or Try agent to start a conversation with it.

Keep refining in the same chat: adjust instructions, choose a model, connect available tools, add supported documents as knowledge, or configure skills and inputs. Follow-up changes update the same agent.

Existing permissions and connector setup still apply. Use Simple mode or the Advanced editor whenever you want the full setup form; see Working with agents for the chat workspace.

Create an agent in Simple mode​

  1. Click Create Agent (or + Create Agent) in the sidebar.
  2. On Create your Agent, describe what you want the agent to do (required). Optionally use Give it a name (optional).
  3. Click Create my Agent.

Devs.ai generates a working first draft—name, greeting, and job description—then opens Simple mode so you can refine it.

Prefer the full form from the start? Use Prefer the full editor? Build it yourself on the create screen.

Refine in Simple mode​

Setup is grouped the way you would explain the agent to a colleague. On small screens, switch between Set up and Try it. On larger screens, the setup accordion sits beside the test chat.

Work through the sections in order (use Next to move on):

Identity​

How your agent introduces itself: avatar, Name, What is it for?, and Greeting (when your organization allows introductions).

Model​

The AI that powers your agent. The default works well for most use cases. Change it when you need a different speed, cost, or capability profile. See Configuring model and data sources for catalog details.

What it does​

Your agent's job description (instructions). Edit the text directly, or describe a change (for example, "Always answer in French") and click Apply with AI. Review the preview, then Use this version or cancel. Use Expand for a larger Write / Preview editor when instructions are long.

What it knows​

Add knowledge the agent should use when answering:

  • Drag and drop files, or click to browse
  • Add link for website URLs (https://…)
  • Import from apps — Google Drive, OneDrive, Jira, Confluence, or GitHub
  • Existing knowledge — reuse a source another agent already uses

Each source shows a status such as Ready, Preparing, or Failed so you can see what actually worked. Remove a source anytime if the agent should stop using it.

What it can do​

Plain-language switches for common abilities. Sensible settings are chosen for you. Typical options include:

  • Search the web
  • Create images (plus an image model when enabled)
  • Analyze spreadsheets
  • Browse websites
  • Workspace — a private working area for files and documents (formerly called Code)

Need API functions, custom inputs, or the full toolbox? Those live in the Advanced editor—use Manage in Advanced when the section points you there.

Connections​

Browse tools your organization already offers, grouped by category (for example CRM, communication, or files). Enable one with a click, or paste an API key when the connection needs one. Some connections show Set up in Advanced when configuration is more complex. Custom MCP servers are set up in the Advanced editor—see Configuring extended capabilities and MCP servers.

Skills​

Link reusable instructions from your library with Add skill so the agent can pull them in when a task calls for them. Create, edit, and import skills in the Advanced editor (Manage in Advanced), or see Skills and Agent skills.

Who can use it​

Choose who can discover and use the agent. You can change this anytime. Typical visibility options include restricted to you, your organization, or anyone with the link—follow the labels in the product.

Try it while you build​

Open Try it (or the Try your Agent pane) to chat with your agent beside the setup.

  1. Make a change in Set up.
  2. Save your changes.
  3. Click Restart chat so the test conversation picks up the latest configuration.

You can hide or show the test chat as you work. Use Open chat in the header when you want a normal conversation outside the builder.

Switch between Simple mode and Advanced editor​

FromControl
Simple modeAdvanced editor in the header
Advanced editorSwitch to Simple mode
Create screen (Simple)Prefer the full editor? Build it yourself

Devs.ai remembers which mode you last chose when you switch explicitly, so create and edit entry points open the surface you prefer. Deep links into a specific Advanced section stay on Advanced so you are not bounced mid-edit.

Advanced editor​

The Advanced editor is the full control panel for the same agent. Use it when you need settings Simple mode does not expose, or when you prefer every field on one form.

To create or open an agent in Advanced:

  1. Click Create Agent and choose the full editor if Simple mode is your default, or open an existing agent and click Advanced editor.
  2. Set name, description, and related profile fields, then save.

Typical Advanced configuration includes:

  • Introduction — the first message the agent displays to users. This sets the tone and helps people understand what the agent can do.

Example (support agent)​

  • Name — Customer Support Agent
  • Description — Helps answer common customer questions using approved documentation and support resources.
  • Instructions — You are a support agent for customers. Answer questions using the provided documentation. Keep responses clear and concise. If information is not available, say so instead of guessing.
  • Example conversation —
    • User: How do I reset my password?
    • Agent: You can reset your password by selecting Forgot password on the sign-in page and following the instructions sent to your email.
⌘J