Back Home

GitHub Repo

AutoGPT 0.11.1 Adds Background Task Continuation; Self-Hosted Deployments Must Migrate in Sync

AutoGPT released 0.11.1 on November 6, adding a mechanism to resume the original conversation after a background subtask completes and changing how results are retrieved and cancellations are propagated. The feature remained behind a flag when merged, and self-hosted teams must complete database and service migrations first.

bisco5 (eBay photograph); Mittelbau camp administration (voucher issuer). · Public domain · Image source
zh-Hant

On November 6, AutoGPT released Platform v0.11.1, adding a mechanism for conversations to resume after background subtasks. When a subtask completes, fails, or stops, the main agent—such as Otto—can pick up the original, idle conversation and continue with the next work. For agent workflows that involve long searches or multiple processing stages, this is more useful than simply returning a completion notice. Release notes

The old implementation set separate time windows for the subtask, follow-up summary, and result delivery. Any follow-up could only produce a summary without using tools. The new implementation removes those separate time limits, while subtasks continue to inherit the deadline of the turn that started them. When the main conversation resumes, it retains that turn’s tool permissions and spending records, and remains subject to the user’s usage limits. Removing fixed wait windows therefore does not remove cost controls. Merged proposal

Results are now handed off through tool-based retrieval: output stays in the sub-conversation, and the main agent retrieves it with get_sub_session_result, avoiding inserting the summary directly into a user message. Stop actions propagate to derived tasks, reducing the risk that background work continues after the main conversation has stopped. From an engineering perspective, the update focuses on preserving execution context and lifecycle management for asynchronous work. Implementation details

Deployment has clear prerequisites. The authors require applying the database migration and deploying compatible versions of the API, database manager, scheduler, and workers before enabling the feature flag; keep it off until then. The flag was also disabled when the change was merged. The authors reported database integration and frontend test results, but did not complete an end-to-end smoke test using a live model provider, so actual long-running execution still needs validation. Deployment and test notes

The release also includes a self-hosted upgrade caveat: default settings no longer include a shared Web Push key, and the backend no longer sends notifications using the old key. After replacing it, existing push subscriptions will stop working. The official self-hosting guide notes that make init-env does not merge new variables into an existing environment file. Engineering teams should compare configuration changes and include push re-subscription, background task cancellation, and result continuation in their upgrade acceptance checks. Release notes, self-hosting guide

Sources

  1. AutoGPT Platform v0.11.1 發布紀錄
  2. 背景子任務接續原對話:PR #16582
  3. Setting Up AutoGPT (Self-Host)