Video Marketer

We have indexed over 1 million TikTok videos for you to explore our Large Visual Memory Model and Agentic QA System.
With this API, you can chat directly with insights from the entire 1-million-video dataset—no need to manually select videos beforehand!

---

Prerequisites

• You have created a memories.ai API key.

---

Host URL

• https://api.memories.ai

Endpoint

POST /serve/api/v1/marketer_chat

---

Request Example

Non-streaming mode

 import requests
import json

headers = {
    "Authorization": "<API_KEY>",
    "Content-Type": "application/json",
}

payload = {
    "prompt": "What does nike recently post?",  # User query
    "session_id": "",  # Chat session ID
    "unique_id": '1'
}

response = requests.post(
    "https://api.memories.ai/serve/api/v1/marketer_chat",
    headers=headers,
    data=json.dumps(payload),
    stream=False
)

print(response.status_code)
print(response.json())

Streaming mode

import requests
import json

headers = {
    "Authorization": "<API_KEY>",
    "Content-Type": "application/json",
    "Accept": "text/event-stream"
}

payload = {
    "prompt": "What does nike recently post?",  # User query
    "session_id": "",  # Chat session ID, Optional
    "unique_id": '<UNIQUE_ID>' # Optional
}

response = requests.post(
    "https://api.memories.ai/serve/api/v1/marketer_chat_stream",
    headers=headers,
    data=json.dumps(payload),
    stream=True
)

if response.status_code != 200:
    print(response.status_code)
    print(response.text)
else:
    try:
        full_chunks = []
        for line in response.iter_lines(decode_unicode=True):
            if not line:
                continue
            lower = line.strip().lower()

            if lower in ('data:"done"', "data:[done]", "data:done", "[done]", "done"):
                print("\n")
                break

            if line.startswith("data:"):
                data = line.replace("data:", "", 1).strip()

                text_piece = None
                try:
                    obj = json.loads(data)
                    if isinstance(obj, dict):
                        for k in ("content", "text", "delta", "message"):
                            if isinstance(obj.get(k), str):
                                text_piece = obj[k]
                                break
                except Exception:
                    pass

                if text_piece is None:
                    text_piece = data

                print(text_piece, end="", flush=True)
                full_chunks.append(text_piece)

        final_text = "".join(full_chunks)
        print("\n=== FINAL ===")
        print(final_text)

    except Exception as e:
        print(str(e))

Request Body

{
  "videoNos": [
    "string"
  ],
  "prompt": "string",
  "session_id": "123456",
  "unique_id": "default",
}

Request Parameters

Name	Location	Type	Required	Description
Authorization	header	string	Yes	API key used for authorization
Accept	header	string	No	text/event-stream for streaming mode
prompt	body	string	Yes	natural language prompt
session_id	body	int	No	ID of the chat session
unique_id	body	string	No	default by default

Response Example

🧠 Thinking Message

{
    "type": "thinking",
    "title": "Based on selected videos, fetch detailed information", 
    "content": "Okay, the user wants a \"Video summary.\" I've been given some selected videos and need to fetch their detailed information to understand their content. This means I need to go",
    "sessionId": "606120397607260160"
}

🔁 Reference Message (ref)

{
	"type": "ref",
	"sessionId": "606143186439766016",
	"ref": [{
		"video": {
			"duration": "10",
			"video_no": "VI606140356924534784",
			"video_name": "test_video_gz_visual_understanding_fuse_s9_video_fuse_4_video_fuse_4"
		},
		"refItems": [{
			"videoNo": "VI606140356924534784",
			"startTime": 23,
			"type": "keyframe"
		}
		{
			"videoNo": "VI606140356924534784",
			"startTime": 30,
			"type": "visual_ts",
			"endTime": 36,
			"text": "A close-up view shows a collection of items on a concrete surface. To the left, two brown, round objects resembling small barrels or planters are visible. One has a blue lid or insert. Next to them, a black corrugated pipe is partially visible, along with a red flexible tube and a yellow pole with a red tip. A grey, textured wall or fence dominates the right side of the frame, with wooden planks visible at the top right."
		},{
			"videoNo": "VI606140356924534784",
			"startTime": 30,
			"type": "audio_ts",
			"endTime": 36,
			"text": "A close-up view shows a collection of items on a concrete surface. To the left, two brown, round objects resembling small barrels or planters are visible. One has a blue lid or insert. Next to them, a black corrugated pipe is partially visible, along with a red flexible tube and a yellow pole with a red tip. A grey, textured wall or fence dominates the right side of the frame, with wooden planks visible at the top right."
		}
]
	}]
}

💬 Content Message

{
  "type": "content",
  "role": "assistant",
  "content": "A\" shape, is suggested to be inspired by Eiffel's past",
  "sessionId": "606122521255088128"
}

✅ Response End Example Success Response (Status Code: 200)

{
  "code": "SUCCESS",
  "data": "Done"
}

⚠️ Error End Conditions Any final response where "data" is not "Done" is considered an error. Common examples include: "data": "Error" "data": "No videos found for the provided video numbers." "data": "user don't login" "data": "Video is not parsed:"

Response End Example

Status code 200

{
  "code": "SUCCESS",
  "data": "Done"
}

Response Result

Status code	Status code msg	Description	Data
200	OK	none	Inline

Response Structure

Status code 200

Name	Type	Required	Restriction	Description
code	string	true	none	Response code
data	object or string	true	none	JSON data. "Done" on success; error message string or structured data otherwise
type	string	true	enum	Type of message, e.g., "thinking", "ref", "content"
title	string	false	none	Title of the thinking message (used when type is "thinking")
content	string	false	none	Text content of the message (used in "thinking" or "content" types)
sessionId	string	true	UUID/ID	ID of the current session
role	string	false	enum	Role of the responder; currently used with "assistant"
ref	array	false	none	List of reference objects containing video and timestamp-based metadata
video	object	true	nested	Video metadata (only when type is "ref")
video_no	string	true	none	Unique video identifier
video_name	string	true	none	Name of the video
duration	string or number	true	seconds	Duration of the video
refItems	array	true	none	Reference annotations such as keyframe, visual_ts, audio_ts
videoNo	string	true	none	Video identifier in refItems (redundant with video_no)
startTime	number	true	seconds	Start timestamp of the referenced segment
endTime	number	optional	seconds	End timestamp (if applicable, e.g., for visual_ts or audio_ts)
text	string	optional	none	Transcribed or described content of the referenced segment