---
title: "September reliability batch"
description: "Telegram posts one live message per run, connected-app logins refresh safely, the workflow compiler rejects unsafe tool steps, and large workflow graphs render faster."
---

> Documentation Index
> Fetch the changelog index at: https://docs.aiagent.app/changelog/llms.txt

# September reliability batch

Telegram posts one live message per run, connected-app logins refresh safely, the workflow compiler rejects unsafe tool steps, and large workflow graphs render faster.

- Telegram live updates post once per run. Repeated messages during a long step are gone.
- Connected-app token refresh is safe under concurrency, so a connection is never revoked by a race.
- The workflow compiler refuses tool steps that would bypass approval in a hand-built graph.
- The workflow builder does a fraction of the work per frame on large graphs, so dragging stays smooth.
- Image generation requests are clamped to the number of images actually billed.

Source: https://docs.aiagent.app/changelog/september-reliability-batch/index.mdx