Agents

Create a reusable AI setup with its own instructions, context, and model, then use it to start focused chats.
5 min read · Reviewed

Create a reusable AI setup with its own instructions, context, and model, then use it to start focused chats.

Understand what an agent saves

An agent packages the setup for a repeatable kind of work. Each agent can save:

  • a name, icon, and description that help people recognize it;
  • instructions that define its job, method, and expected output;
  • Bidlo pages, records, or files for it to reference; and
  • the model and reasoning level it should use.

For example, you could create an agent that reviews project information, identifies missing bid details, and always returns the same checklist format. You would supply the current project when you start each chat.

Create an agent

  1. From Home, select Chats in the left sidebar.
  2. In the Chats panel, find Agents and select New.
  3. In New agent, enter a short, recognizable Name.
  4. Optional: select Change agent icon to choose a different icon and color.
  5. Add a Description that tells teammates when to use the agent.
  6. In Instructions, describe the agent's role, the steps it should follow, and the format it should return.
  7. Under Context, select Add information for your agent to reference. Search for a Bidlo page or record and select it, or choose Upload new file.
  8. Under Model, select Choose model. Use the provider tabs or search box to choose a model, then set its reasoning level if that option is available.
  9. In Preview, enter a representative request to check how the agent responds.
  10. When the setup produces the result you want, select Save.

The saved agent appears under Agents in the Chats panel for the current team.

Write instructions that are easy to reuse

Strong instructions answer four questions:

  • Role: What kind of assistant is this?
  • Task: What should it do each time?
  • Method: What steps or checks should it follow?
  • Output: What should the finished answer look like?

For example:

Review the supplied project context for bid readiness. Identify missing due dates, owner requirements, and scope details. Return a table with the issue, supporting source, and recommended follow-up. Do not guess when a detail is missing.

Use the Preview panel to test a normal request, an incomplete request, and an edge case before saving. If the output varies too much, make the method or output format more explicit in Instructions.

Add useful context

Saved context gives the agent reference material it should use across chats. In the context picker, you can search the current Bidlo workspace for pages and records or select Upload new file.

Use saved context for stable material such as a standard operating procedure, approved template, or reference page. Supply changing material, such as the project being reviewed today, in the chat where you run the agent.

Choose a model

Select Choose model to browse available providers, search for a model, and set a reasoning level. The menu can include provider tabs such as OpenAI, Anthropic, xAI, Google, and Z.AI; the exact models available can change.

Keep the default when you don't have a specific model requirement. Change the model or reasoning level when the agent's task calls for a different balance of speed and depth, then compare the result in Preview before saving.

Start a chat with an agent

  1. Select Chats in the left sidebar.
  2. In the Chats panel, select the saved agent under Agents.
  3. Enter the current request in the chat prompt.
  4. Add or mention any page, record, or file that is specific to this run.
  5. Send the request and review the result before using it in your work.

The new chat uses the agent's saved instructions, context, model, and reasoning setting. Your prompt supplies the goal and details for this particular run.

Improve an agent's results

If an agent's answer is incomplete or inconsistent, check the setup in this order:

  1. Make the requested outcome more specific in the current chat.
  2. Add the missing page, record, or file as context.
  3. Clarify the required checks or output format in the agent's instructions.
  4. Test the revised setup in Preview with the same representative request.
  5. Save only after the preview returns the structure and level of detail you expect.

For high-impact work, verify the source records and any generated files before approving, sharing, or submitting the result.

Agents and skills serve different purposes

Use an agent when you want a saved AI configuration that starts a focused chat with its own context and model. Use a skill when you want to insert a reusable set of team instructions into another AI request.

FAQs

When should I create an agent instead of starting a normal chat?

Create an agent when the same role, instructions, tools, and context should be reused. Use a normal chat for a one-time request.

What should I change when an agent’s results are inconsistent?

Tighten the instructions, add the missing source context, and define the expected output and boundaries more precisely. Test the revised agent on a representative task.