Configure Agent Prompt
Agent instructions define how the AI assistant forms replies: role, style, escalation rules, and output format. In Fast Line Pro you edit them in the Instructions block on the agent card, choosing an editor mode for simple or structured prompts.
When Needed
- You need to set the agent's response style (formal, informal, technical).
- You want to describe role, behavior rules, and escalation for customers.
- You need JSON output schemas, XML sections, or tool definitions in the prompt.
- You want to personalize replies with variables in the
{{variable_name}}format.
What's Important to Know
- Agent instructions — the system prompt that defines AI assistant behavior.
- The editor has three modes: Visual, Raw, and Monaco.
- For prompts with XML tags, JSON blocks, or output schemas, use Monaco or Raw — Visual is disabled for such text because the visual editor can corrupt markup on save.
- Changes apply to live dialogs only after you click Publish on the agent card.
Prerequisites
Instructions Editor Modes
In the Instructions block header, switch editor modes:
| Mode | When to use |
|---|---|
| Visual | Simple prompts with headings, lists, and emphasis — no XML tags or JSON blocks. Markdown editing in visual mode. |
| Raw | The same text as Visual mode, in a plain textarea with monospace font. Handy for precise copy-paste or edits without syntax highlighting. |
| Monaco | Structured prompts: XML sections (<role>, <tools>), ```json blocks, Markdown tables, {{name}} variables. Syntax highlighting and comfortable editing of long text. |
Automatic Mode Selection
If the prompt contains:
- XML tags (for example,
<role>), - code blocks
```jsonor```xml, - a raw JSON object,
the system disables Visual and opens Monaco. You can switch to Raw manually at any time.
Recommended Prompt Format
For complex agents (tools, routers, response contracts), combine Markdown + XML + JSON:
| Element | Purpose | Example |
|---|---|---|
| Markdown headings | Section structure | # Role, ## Rules |
| Lists and bold | Rules and limits | - Do not invent facts |
| Markdown tables | Parameters, comparisons | | Channel | SLA | |
| XML sections | Logical prompt blocks | <role>, <primary_objective>, <tools>, <escalation_policy>, <language_rules> |
```json blocks |
Output schema, routing | JSON response fields |
| Variables | Data from scenarios | {{client_name}}, {{company_name}} |
Use variables in instructions only as {{variable_name}} (double curly braces). Names must match agent variables and action parameters in your scenario. See Configure agent variables.
Short Structure Example
# Support consultant
<role>
You are an AI support assistant for {{company_name}}.
</role>
<primary_objective>
Answer from the knowledge base. If there is no answer — escalate to an operator.
</primary_objective>
<language_rules>
- Language: English
- Tone: friendly, professional
</language_rules>
```json
{
"reply": "string",
"needs_escalation": "boolean"
}
```
For simple FAQ prompts, Visual mode without XML and JSON is enough.
Step-by-Step Instructions
- In Fast Line Pro, open the Agents tab and click the agent name.
- Find the Instructions block.
- Choose an editor mode:
- simple text → Visual or Raw;
- XML, JSON, tables, long structured prompt → Monaco (or wait for Monaco to open automatically if the text already contains structure).
- Enter or edit instructions. Add variables as
{{variable_name}}when needed. - If AI assistant is enabled in your system, click Ask AI in the block header. The assistant opens with a prefill about Markdown, JSON/XML, and Monaco mode.
- Review the text in the chosen mode (for structured prompts — in Monaco or Raw).
- Click Publish at the top of the agent card — instructions are saved and applied to future dialogs.
Monaco: What You See in the UI
- Syntax highlighting for headings, XML tags, inline code,
```json/```xmlblocks, bold text, and tables. - Code and tables use a monospace font; inline code is colored without a pink background.
- The editor grows with content up to a maximum height, then scrolls internally; the page scrolls for sections below.
What Happens After
Instructions are saved on the agent card. After Publish, the agent uses the new prompt in dialogs. Until you publish, live conversations keep the previous version.
How to Verify It Worked
- Instruction text displays correctly in the chosen mode (especially XML and JSON in Monaco).
- After Publish, the card shows no unsaved changes.
- A test dialog in your scenario follows the prompt rules (tone, escalation, format).