how to get chat history when we use realtime model like geminai?

Thanks for reaching out, Rahul. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

what i wnat i s, i created gemini relatime aget, i wnat to store the chat history to some external service , how to do it??

Thanks — you want to capture chat history from your Gemini realtime agent and store it in an external service. You can do this by listening to the AgentSession events and forwarding each conversation item to your storage (DB, HTTP endpoint, message queue, etc.). Below are the concise steps and a ready-to-adapt example. What I’m addressing: storing chat history externally from a Gemini realtime AgentSession. Steps • Subscribe to the session’s conversation events (use

conversation_item_added

to capture each message). • Extract role, text, timestamps, and any metadata from the event’s

ChatItem

. • Send that payload to your external service (HTTP POST, DB insert, publish to queue). • Optionally persist incremental transcripts as they arrive (for near-real-time storage) and reconcile with final messages on turn completion. Python example (async):

from livekit.agents import AgentSession
import aiohttp

session = AgentSession(...)  # your realtime Gemini agent session

@session.on("conversation_item_added")
async def on_conversation_item(item):
    role = item.message.role
    timestamp = item.message.created_at
    content = ""
    for part in item.message.content:
        if isinstance(part, str):
            content += part
        elif hasattr(part, "text"):
            content += part.text

    payload = {"role": role, "text": content, "timestamp": timestamp}

    async with aiohttp.ClientSession() as http:
        await <http://http.post|http.post>("<https://your-external-service.example.com/chat>", json=payload)

Node/TypeScript sketch:

const session = new AgentSession(...);

session.on("conversation_item_added", async (item) => {
  const role = item.message.role;
  const timestamp = item.message.created_at;
  let text = "";
  for (const part of item.message.content) {
    if (typeof part === "string") text += part;
    else if (part.text) text += part.text;
  }
  await fetch("<https://your-external-service.example.com/chat>", {
    method: "POST",
    headers: {"Content-Type":"application/json"},
    body: JSON.stringify({ role, text, timestamp })
  });
});

Notes & best practices • The event includes user and assistant messages and metadata (role, timestamps). • If you need audio or richer multimodal data, include relevant fields from the

ChatItem

(images, attachments) when available. • If you must reload history later, store text-based transcripts; realtime models only support loading history in text form and loading large histories can affect audio behavior. • To avoid duplicates, include a unique message ID or use the item’s created timestamp when upserting into your storage. If you tell me which SDK/language and which external service (Postgres, Firebase, S3, HTTP API, etc.) you plan to use, I’ll give a tailored example and schema. Sources: Text and transcriptions | LiveKit docs | Agent session | LiveKit docs | How to capture real-time agent transcriptions during conversations | Agent speech and audio | LiveKit docs

can we do it form session .history on room closing??

Yes — you can read

session.history

when the room/session is closing and then store it externally. What I’m addressing: retrieving full chat history at shutdown and persisting it. How to do it (summary) • Add a shutdown callback or listen for your session/room-close event. • In that handler, iterate

session.history

(contains items with role, text, timestamps, metadata). • Send the assembled payload to your external service (HTTP endpoint, DB insert, S3, etc.) and handle retries/upsserts to avoid data loss. Python shutdown example:

async def on_shutdown():
    records = []
    for item in session.history:
        records.append({
            "id": item.message.id,
            "role": item.message.role,
            "text": "".join(part if isinstance(part, str) else getattr(part, "text", "") for part in item.message.content),
            "timestamp": item.message.created_at,
        })
    # send `records` to your external service
    await send_to_external_service(records)

ctx.add_shutdown_callback(on_shutdown)

Notes and best practices • session.history contains the complete transcript up to close — use it for a final, authoritative export. • Include message IDs or timestamps to avoid duplicates when upserting. • If you need near-real-time backup too, combine this final export with streaming writes from

conversation_item_added

. • If you require audio/multimodal artifacts, make sure you also persist those files/refs during the session (history is text-first). Tell me which SDK/language and which external storage you’ll use (Postgres, Firebase, S3, HTTP API, etc.), and I’ll provide a concrete example and schema. Sources: Agents and handoffs | LiveKit docs | Collect recording consent with tasks | LiveKit docs

Error extracting chat history: 'ChatContext' object is not iterable

Thanks — I’m responding to the "'ChatContext' object is not iterable" error you reported when extracting history at session/room close. Quick answer: don’t iterate the ChatContext directly. Use its

.messages

list (or

text_content()

helpers) instead. Then serialize/send that list to your external service in your shutdown handler. If you want, tell me: • which SDK/language (Python or Node/TS), and • the code snippet where the error occurs (the handler you added for room close), and I’ll give a corrected, copy-paste example that writes

session.history.messages

to your storage (HTTP, Postgres, S3, etc.). Sources: Text and transcriptions | LiveKit docs | Agents and handoffs | LiveKit docs | Vision | LiveKit docs