Skip to main content
POST
/
api
/
v2
/
agents
/
{agent_id}
/
answer
curl -X POST "http://localhost:8000/api/v2/agents/my-agent/answer" \
  -H "X-Session-Token: your_session_token" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "How should we contact the user?",
    "limit": 5,
    "temperature": 0.7,
    "ai_model": "anthropic.claude-sonnet-4-6",
    "kiosk_mode": false
  }'

{
  "agent_id": "my-agent",
  "session_id": "sess_123abc",
  "question": "How should we contact the user?",
  "answer": "Based on stored preferences, the user prefers email communication. Send email during business hours.",
  "sources": [
    {
      "id": "3e681f12-a28c-4d1d-9632-b8dadf1f9d0c",
      "score": 0.86
    }
  ],
  "namespace": "memanto_agent_my-agent"
}

​
Overview

Answer questions using stored agent memories. This operation retrieves relevant context from the agent’s Moorcheh namespace (scoped by the active session) and calls Moorcheh answer generation to produce a grounded reply.

​
Authentication

API clients do not send an API key or Authorization header.
​
X-Session-Token
string
required
Session token from Activate Agent. Must match agent_id.
​
Content-Type
string
required
Must be application/json

​
Path Parameters

​
agent_id
string
required
The unique identifier of the agent.

​
Body

​
question
string
required
The question to answer using retrieved memories as context.
​
limit
integer
Maximum memories to use as context (top_k). Range 1–100. If omitted, the server default applies (see deployment configuration).
​
temperature
number
LLM temperature, 0.0–2.0. If omitted, the server default applies.
​
ai_model
string
Model identifier for answer generation (snake_case field name: ai_model). If omitted, the server default applies.
​
kiosk_mode
boolean
When true, relevance filtering uses a confidence threshold. When false (default), threshold is ignored and not sent to Moorcheh.
​
threshold
number
Confidence threshold (0.0–1.0). Only used when kiosk_mode is true. If kiosk_mode is true and threshold is omitted, the server uses 0.15.
curl -X POST "http://localhost:8000/api/v2/agents/my-agent/answer" \
  -H "X-Session-Token: your_session_token" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "How should we contact the user?",
    "limit": 5,
    "temperature": 0.7,
    "ai_model": "anthropic.claude-sonnet-4-6",
    "kiosk_mode": false
  }'

{
  "agent_id": "my-agent",
  "session_id": "sess_123abc",
  "question": "How should we contact the user?",
  "answer": "Based on stored preferences, the user prefers email communication. Send email during business hours.",
  "sources": [
    {
      "id": "3e681f12-a28c-4d1d-9632-b8dadf1f9d0c",
      "score": 0.86
    }
  ],
  "namespace": "memanto_agent_my-agent"
}

​
Temperature Guide

  • 0.0-0.5: Conservative, factual responses - best for technical documentation
  • 0.5-1.0: Balanced creativity - good for general Q&A
  • 1.0-2.0: More creative and varied responses - use carefully for factual content

​
Available Models

Model IDNameProviderDescriptionCredits
anthropic.claude-sonnet-4-6Claude Sonnet 4.6AnthropicFast flagship: coding, tools, long docs and RAG (~1M context).3
anthropic.claude-opus-4-6-v1Claude Opus 4.6AnthropicDeepest reasoning and hardest tasks; pick when quality matters most (~1M context).3
meta.llama4-maverick-17b-instruct-v1:0Llama 4 Maverick 17BMetaLong context, summarization, function calling, fine-tuning friendly.3
amazon.nova-pro-v1:0Amazon Nova ProAmazonChat, math, and structured answers for AWS-style workloads.2
deepseek.r1-v1:0DeepSeek R1DeepSeekStep-by-step reasoning; math, logic, and technical explanations.1
deepseek.v3.2DeepSeek V3.2DeepSeekEfficient general Q&A, multilingual, everyday RAG (~164K context).2
openai.gpt-oss-120b-1:0OpenAI GPT OSS 120BOpenAILarge generalist: research-style answers and long-form writing.3
qwen.qwen3-32b-v1:0Qwen 3 32BQwenCode and bilingual (EN/ZH) tasks in a smaller footprint.2
qwen.qwen3-next-80b-a3bQwen3 Next 80B A3BQwenMoE model for long chats, docs, and code at scale (~256K context).1

​
Next Steps

  • Recall to fetch raw memories without generating an AI answer
  • Remember to add more context to the agent