AMADEV Agent Validation
Use amallm-agent-validation.json as the canonical first-turn payload for /v1/chat/completions.
Validation flow:
GET /v1/modelsGET /v1/toolsPOST /v1/chat/completionswithamallm-agent-validation.json- If the selected model reports
supports_tools: true, expect:choices[0].message.tool_callschoices[0].finish_reason === "tool_calls"
- Append the returned assistant message and a tool result message:
Code
{
"role": "tool",
"tool_call_id": "`<id-from-turn-1>`",
"name": "get_weather",
"content": "{\"city\":\"Bangkok\",\"temperature_c\":33,\"condition\":\"Cloudy\"}"
}
- Submit the second turn to
/v1/chat/completions - Expect:
- final assistant
content choices[0].finish_reason === "stop"
- final assistant
Current compatibility behavior:
- If
/v1/modelsreportssupports_tools: false, AMADEV will reject agentic tool-calling requests with a structured compatibility error instead of silently downgrading them into plain chat. /v1/toolsand/v1/tools/executeremain available for discovery and direct execution even when canonical chat tool-calling is disabled.
