Skip to main content
Field reference: examples — every field, type, default, and tier for this page.
Examples are sample situation-and-response pairs that demonstrate how your agent should behave. They guide tone, response length, format, and domain focus — without needing to write lengthy instructions.

​
Adding an example

  1. Open your agent and go to Examples in the sidebar.
  2. Click Add example, give it a name, and Create.
  3. Fill in the three fields on the example card:
    • Situation / Input — what the user says or asks.
    • Agent Action / Output — how the agent should respond.
    • Reasoning (optional) — why this is the right decision. Useful for capturing intent; the agent sees this alongside the input/output.
  4. Save Changes.
Examples are included in the agent’s context at inference time, so the LLM can learn from them directly.

​
Worked examples

Hello world — set the tone. Teach the agent to be concise:
  • Situation: “How do I reset my password?”
  • Output: “Go to Settings → Security → Reset password, then check your email for the link. Anything else?”
  • Reasoning: Short, actionable, ends with an offer to help further.
One pair like this, placed first, nudges every reply toward the same crisp style. Medium — enforce a format. Teach the agent to answer comparisons as a table:
  • Situation: “What’s the difference between the Pro and Team plans?”
  • Output: (a small markdown table comparing the two plans across price, seats, and features)
  • Reasoning: Comparisons are clearer as a table than prose; always use this format for “difference between X and Y” questions.
Complex — capture a judgement call. Teach the agent how to handle a tricky edge case:
  • Situation: “Can you give me a discount? I’m a long-time customer.”
  • Output: “I can’t apply discounts directly, but I can flag your account for our retention team who handle loyalty offers — want me to do that?”
  • Reasoning: Never invent discounts (a policy boundary), but don’t just say no — offer the compliant path. This pattern generalises to any “can you do X that you’re not allowed to do” request.
The Reasoning field is where examples earn their keep on hard cases: it tells the agent not just what to do but why, so it generalises the principle to similar situations rather than parroting the exact wording.

​
When to use examples

Examples are most useful when:
  • The agent needs to match a specific tone (formal, casual, concise).
  • Responses require a particular format (bullet lists, numbered steps, tables).
  • The domain is specialised and general LLM behaviour is too generic.
  • You want consistent handling of common questions.

​
Ordering

Examples are presented to the LLM in the order they appear in the list. Drag a card by its header to reorder; the new order saves automatically. Put your most representative examples first. Each card carries an [EXAMPLE-N] badge — the example’s citation ID, matching its 1-based position in the list. The agent uses that ID to reference the example in its reasoning traces, so a card’s ID shifts when you reorder.

​
Limits

Each agent supports up to 1000 examples. The Situation, Output, and Reasoning fields each share the same character limit as other text fields.

​
Importing examples

Use the Upload Example Definition button to bulk-load examples from a file or pasted text. The upload page accepts Markdown, JSON, YAML, and TOML (.md, .json, .yaml, .yml, .toml) — pick a format tab there for a ready-to-copy template. Each entry should include a name, input, output, and optional reasoning. Uploaded examples are added to your existing list.

​
Manage via the Management MCP

This page can be configured programmatically through the Management MCP — useful for AI agents and CI pipelines. Example:
Partial updates are supported — only the fields you include are changed. Call config.get to read the current values and schema.get (or the HTTP schema URL above) to see the full field list.