Agent IA
Runs an autonomous AI agent with a goal (task). The agent can use tools, perform multiple iterations, and return a synthesized response. The main field is answer (not content).
Parameters
| Parameter | Type | Required | Variable | Description |
|---|---|---|---|---|
task | dynamic value | Yes | Yes | Goal or instruction for the autonomous agent (min 10 characters). (min 10 chars) |
systemPrompt | text | No | Yes | System instructions defining the agent behavior and constraints. |
model.provider | text | Yes | No | AI model provider for reasoning (e.g. openai, anthropic). |
model.model | text | Yes | No | AI model identifier to use (e.g. gpt-4o, claude-3-opus). |
parameters.temperature | number | No | No | Agent reasoning creativity (0 = deterministic, 1 = creative). (Default: 0.5, min 0, max 2) |
parameters.maxTokens | number | No | No | Maximum number of tokens per agent iteration. (Default: 4096, min 1, max 128000) |
enabledTools | array | No | No | List of tools available to the agent (semantic_query, http_request, etc.). |
maxIterations | number | No | No | Maximum iterations before forced synthesis (safety limit). (Default: 10, min 1, max 50) |
contextVariables | array | No | No | Context variables to pass to the agent for its session. |
enableSubAgents | boolean | No | No | Enable delegation to sub-agents for complex tasks. (Default: false) |
outputVariable | text | No | No | Output variable name containing the agent final result. |
Parameters marked Variable = Yes accept the
{{blockName.field}}syntax.
Output
Output variable : agentResult
{
"success": false,
"answer": "...",
"toolResults": "...",
"stats": {},
"tokensInput": 0,
"tokensOutput": 0,
"totalTokens": 0,
"sessionId": "..."
}
Example
Analyze a dataset and produce a report.
Input :
{"task": "Analyse les ventes Q1 2026"}
Output :
{"success": true, "answer": "Les ventes ont augmente de 15% au Q1 2026...", "toolResults": [], "stats": {"iterations": 2, "toolCalls": 0, "durationMs": 3200}, "tokensInput": 150, "tokensOutput": 85, "totalTokens": 235}
Tip
Use {{agentResult.answer}} for the response (not {{agentResult.content}}). The task field is required (min 10 characters), not goal.