If you use Telegram to talk with your agents, this release makes the experience noticeably more reliable. Messages arrive in order, responses do not get lost, and duplicate notifications are a thing of the past.
Serialized Message Processing
When you send multiple messages quickly, the system previously allowed them to be processed concurrently. This could cause overlapping AI responses, duplicated replies, or messages arriving out of order.
Messages are now processed one at a time per conversation. If a message arrives while another is still being handled, it waits in a queue rather than racing ahead. If the wait exceeds a reasonable timeout, you receive a brief "still thinking" notification so you know the system has not dropped your message.
This serialization works correctly even when your messages are handled by different server instances — it is not limited to a single process.
Guaranteed Message Delivery
A subtle timing issue previously caused some Telegram messages to be lost: the system would consider a response "done" before the Telegram send had actually completed. This has been fixed — responses are now fully delivered before the processing pipeline moves on.
Entity-Level Notification Cooldown
When the same planning entity is re-executed across multiple runs in quick succession (common in iterative dev loops), the system now suppresses duplicate Telegram notifications within a five-minute window. You still get notified the first time — but repeated runs of the same entity will not flood your Telegram with identical prompts.
Delivery Tracking
Each Telegram notification now tracks its delivery status (pending, sent, or failed). Failed deliveries no longer permanently block retries — the system can reattempt them on subsequent runs instead of assuming the notification was already sent.