Notifications
Clear all
Topic starter
Hello everyone! I'm developing an AI consultant within Telegram and have encountered a slight hurdle with the workflow implementation.
My Current Task: I need to collect user messages for approximately 10-15 seconds before forwarding them to the AI model. The goals are to:
- Ensure comprehensive conversation context is gathered.
- Decrease the frequency of API calls.
- Enhance the quality of responses.
What I’ve Already Tried:
- Utilized the Telegram Trigger.
- Planned to use OpenAI/OpenRouter for generating responses.
- Intend to implement temporary message storage within the workflow.
Questions for the Community:
- What is the recommended method for implementing message buffering in callin.io?
- What are the best practices for managing sequential messages?
- Are there any existing templates or examples available for this type of workflow?
I would greatly appreciate any advice or guidance!
My Tech Stack:
- callin.io
- Telegram
- OpenAI/OpenRouter
Posted : 12/07/2025 10:50 am
What you’re describing (waiting 10–15s before sending messages to the model) isn’t natively supported in callin.io. The default flow is always “event → action → response” with no buffering.
Options that work today
-
Make or Zapier
-
Connect Telegram as the trigger.
-
Add a Delay of 10–15 seconds.
-
While the delay runs, store incoming messages in Storage (Make) or Storage by Zapier.
-
Once the delay finishes, concatenate all stored messages and send them to your model (OpenAI/OpenRouter).
-
-
Buffer management
-
Use the Telegram
chat_idas the storage key. -
Each new message gets appended to that list.
-
When the delay is up, fetch the list, clear it, and forward it to the model.
-
For clarity, add separators like “---” between messages so the model understands they’re sequential.
-
-
Best practices
-
Define an inactivity window (e.g., if >15s without new messages, flush the buffer).
-
A 10–15s window usually balances context gathering with acceptable latency.
-
Don’t concatenate endless text—keep it to coherent chunks for better model responses.
-
Posted : 19/09/2025 8:08 am