MR AI RAG — Memory Chatbot API Documentation

🧠 API Reference Documentation

MR AI RAG
Memory Chatbot API

Complete reference for embedding AI memory chatbots, managing conversations, and measuring voice interaction latency.

Version

v2.0

Base URL

https://test.3rdai.co

Auth

API Key or Client Token

Format

JSON (REST)

Table of Contents

1 Authentication

2 Memory Management

3 Chatbot Conversations

4 Source Uploads

5 Reels & Video Generation

6 Voice Latency Guide

7 Embed & QR Code

8 Error Codes

1 Authentication

All API calls require an API key in the X-API-Key header, or a client session token in X-Client-Token. Obtain API keys from Dashboard → Settings → API Keys.

Note: Client endpoints require X-Client-Token. Admin endpoints require X-Admin-Token.

POST/api/clients/loginEmail + password login

Field	Type	Required	Description
email	string	Required	User email address
password	string	Required	Account password

Response 200

{ "client_id": "clt-...", "token": "clt-...", "name": "John Doe", "email": "user@example.com" }

POST/api/clients/google-loginGoogle OAuth login

Field	Type	Required	Description
id_token	string	Required	Google ID token from Google Sign-In flow

2 Memory Management

A Memory is an AI knowledge base that stores uploaded documents, websites, and videos. Each chatbot is backed by a distinct memory.

GET/api/memoryList all memories

Header: X-Client-Token: {token}

POST/api/memoryCreate a new memory

Field	Type	Required	Description
name	string	Required	Chatbot memory name
description	string	Optional	Short description
provider	string	Optional	"gemini" (default) \| "openai"

DELETE/api/memory/{memory_id}Delete a memory (irreversible)

All sources and chat history will be permanently deleted.

3 Chatbot Conversations

Send questions to a memory chatbot. The AI answers using only the knowledge stored in that memory.

POST/api/memory/{memory_id}/askAsk the chatbot a question

Field	Type	Required	Description
question	string	Required	User's question
top_k	integer	Optional	Number of chunks to retrieve (default: 5)
history	array	Optional	Chat history e.g. [{"role": "user", "content": "..."}]

Response 200

{
  "answer": "Based on the knowledge base, ...",
  "sources": ["document.pdf", "website.com"],
  "latency_ms": 1240,
  "history_saved": true
}

GET/api/memory/{memory_id}/chat-historyGet Q&A pairs

Query Param	Type	Default	Description
limit	integer	10	Max Q&A pairs to return
offset	integer	0	Pagination offset

4 Source Uploads

Add knowledge to a memory by uploading PDFs, pasting web URLs, or providing YouTube links.

POST/api/memory/{memory_id}/upload-pdfUpload a PDF document

Content-Type: multipart/form-data | Form field: file

POST/api/memory/{memory_id}/ingest-urlAdd a web page to memory

Field	Type	Description
url	string	HTTP/HTTPS URL of the article/website

POST/api/memory/{memory_id}/ingest-youtubeAdd a YouTube video to memory

Field	Type	Description
url	string	YouTube video link

POST/api/memory/{memory_id}/ingest-json-urlFetch and index a JSON API/file from URL

Field	Type	Description
url	string	HTTP/HTTPS URL of the JSON data
title	string	Descriptive title for the source

POST/api/memory/{memory_id}/ingest-mongodbConnect to MongoDB & index documents (Live Real-Time)

Field	Type	Required	Description
connection_string	string	Required	MongoDB URI (mongodb+srv://...)
database	string	Required	MongoDB Database Name
collection	string	Required	MongoDB Collection Name
title	string	Optional	Descriptive title for the source

5 Reels & Video Generation

Generate short video reels from memory content or a specific answer. Placeholder videos are returned — integrate D-ID, HeyGen, or Runway ML for production use.

POST/api/memory/{memory_id}/generate-reelGenerate a video reel

Field	Type	Required	Description
topic	string	Optional	Topic/title for the reel (default: memory name)
style	string	Optional	"cinematic" \| "minimal" \| "energetic"
content_text	string	Optional	Specific answer text to generate from (max 5000 chars)

Response 200

{ "success": true, "video_url": "https://...", "topic": "...", "style": "cinematic" }

6 Voice Latency Guide

Voice latency is the delay between a user speaking and the chatbot starting to respond. The total is the sum of several pipeline stages:

Component	Typical	Range	Rating
Speech-to-Text (STT) Transcribing voice to text (e.g. Whisper, Google STT)	200–400ms	100–800ms	Fast
Vector Search (RAG) Finding relevant chunks in the memory	50–150ms	30–400ms	Fast
LLM Generation — Gemini Generating the AI answer (Google Gemini)	800–1500ms	500–3000ms	Moderate
LLM Generation — GPT-4o Generating the AI answer (OpenAI)	600–1200ms	400–2500ms	Moderate
Text-to-Speech (TTS) Converting answer text to audio	300–600ms	200–1200ms	Moderate
Network Round-trip API calls + server overhead	50–200ms	20–500ms	Fast
Total — Non-streaming	1.5–3.0s	1.0–5.0s	Acceptable
Total — Streaming (first token)	300–700ms	200ms–1.5s	Excellent

Tip: Enable streaming ("stream": true in the chat request) to reduce perceived latency. The first token appears in ~300–700ms, making the response feel instant to users.

Benchmarks by Scenario

Scenario	Total Latency	Notes
Simple Q&A (1–2 sources)	~800ms	Streaming + small context
Complex multi-document query (10+ sources)	~2.5s	More chunks to process
Cold start / first message	~1.5–2s	Model warm-up overhead
Subsequent messages (warm)	~600ms–1.2s	Embedded context cache
Full voice pipeline (STT + LLM + TTS)	~2.0–4.0s	End-to-end voice interaction

7 Embed & QR Code

Embed the chatbot on any website using an iframe, or share a QR code that opens the public chat interface.

Embed iframe

<!-- Add to any webpage -->
<iframe
  src="https://your-domain.com/memory-chat-public?id={memory_id}"
  width="400" height="600"
  style="border:none;border-radius:16px"
  allow="microphone"
></iframe>

Public QR URL

https://your-domain.com/memory-chat-public?id={memory_id}

8 Error Codes

Code	Error	Cause	Fix
400	Bad Request	Missing or invalid body fields	Check required parameters
401	Unauthorized	Missing or invalid API key / token	Check X-API-Key or X-Client-Token header
403	Forbidden	Accessing another user's resource	Use the correct client token
404	Not Found	Memory or resource doesn't exist	Verify the memory_id or resource path
422	Validation Error	Field type or constraint mismatch	Check field types (int vs string)
500	Server Error	Unexpected server-side failure	Check server logs; retry after moment