Why Telegram AI replies are limited to private chats
Groups and channels involve more participants and more ambiguous context, so one incorrect reply can affect an entire community. Owenrgram restricts AI auto reply triggers to private messages and does not call a model simply because a keyword appears in a group.
Bot chats and attachment-only messages are excluded as well. This boundary is enforced by product logic rather than relying on a prompt that repeatedly asks the model not to answer groups.
- Process ordinary user private chats only
- Never respond automatically in groups or channels
- Bot chats and attachment-only messages do not call the model
Configure the endpoint, model, and token per account
Each account can use an OpenAI-compatible API endpoint, model name, and token. Owenrgram does not require one model provider or force every account to share one global model configuration.
Model requests still travel to the third-party service selected by the user. Pricing, data processing, retention, and availability remain subject to that provider, so operators should review its privacy and logging terms before configuration.
- Custom OpenAI-compatible API endpoint
- Account-level model and token configuration
- Third-party model cost and policy remain provider-controlled
Reply from recent conversation context
Automatic replies can include the most recent 1–50 messages from both sides. This helps the model understand the previous question, the answer already given, and the issue currently being handled instead of generating from one isolated message.
A larger context may improve continuity, but it also increases token use and the amount of conversation data sent to the model provider. Operators choose the window according to quality, cost, and privacy requirements.
- Choose a context window from 1 to 50 messages
- Include recent messages from both participants
- Larger windows increase token use and data exposure
AI generation still follows account and platform boundaries
Generated text does not mean Telegram accepted the message. Delivery still runs through the account Runtime and the final Operation state determines success. Permission failures, Flood Wait, and network errors remain visible.
AI replies fit explicit, repetitive private-chat scenarios, but should not replace human judgement for sensitive conversations. The feature is account-scoped and can be disabled whenever an operator needs to take back control.
- Generation and delivery are separate steps
- Final Operation state confirms the outcome
- Disable automatic replies per account
Frequently asked questions
Will AI reply in Telegram groups or channels?
No. Automatic replies process private chats only; groups and channels never trigger a model call.
Can I use my own model API and token?
Yes. Each account can configure an OpenAI-compatible endpoint, model, and token.
How much conversation context can an AI reply use?
You can choose the most recent 1–50 messages from both sides.
Do image-only or file-only messages trigger AI?
No. Attachment-only messages do not currently trigger a model call.