> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bellows.cloud/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> What changed in each Bellows release, in plain words.

User-facing changes only. Self-hosting upgrade notes live in the release notes on [GitHub](https://github.com/Bellows-AI/bellows/releases).

**v0.0.15**

2026-11-06

**New**

* **MCP server.** Drive Bellows from Claude Code, Claude.ai or any MCP client. It is on for Bellows Cloud and switched on by the admin on self-hosted installs. See [Connect an MCP client](https://docs.bellows.cloud/mcp/connect.md).
* **Task titles.** Every task has a short title you can edit from the task page.
* **Base branch.** Pick the branch a task starts from in the composer.
* **Skill menu.** Type `/` in the composer to pick a [skill](https://docs.bellows.cloud/guides/skills.md).

**Improved**

* A redesigned task page: controls on top, published work in an **Outcome** tile, and collapsible sections.
* New task and follow-up share one chat input, with attach and send inside the box.
* A run that changed nothing skips its gates or reuses the last result.

**Fixed**

* Recovery after an infrastructure loss keeps the agent's own branch.

**v0.0.14**

2026-11-02

**New**

* **Notifications.** Get told when a task is ready for review, blocked on a person, or has a question. Discord is supported first. See [Connections](https://docs.bellows.cloud/guides/connections.md#discord-notifications).
* **Attachments** reach the agent, with image previews and inline tiles in the chat. See [Attachments](https://docs.bellows.cloud/guides/attachments.md).
* **Connect block.** A task that needs a connection asks for it, then resumes once you connect.
* **Admin actions.** Admins can follow up on, retry or control the wait of another member's task. These actions are audited and the member is notified.

**Improved**

* A compact layout for Settings, notifications, repositories, executors, environment and the task composer, and a new integrations overview and catalog.
* Task detail shows live gate progress at the end of a run.
* Environment secrets use one stored format.

**v0.0.13**

2026-10-30

**New**

* **Managed connections** with personal defaults, and a shared sign-in flow behind them.
* **Jira through Atlassian sign-in.** Connect Jira with your own Atlassian account. Older API-token Jira connections were disconnected, so reconnect once. See [Connections](https://docs.bellows.cloud/guides/connections.md#jira).

**Changed**

* Environment secrets are now stored encrypted. See [Environment variables and secrets](https://docs.bellows.cloud/guides/env-and-secrets.md).
* The agent may use any git and `gh` operation in its task checkout.

**Fixed**

* A leftover runner stuck terminating now hands the task back as blocked instead of letting it expire.
* The gates skill no longer asks for a stale tool.

**v0.0.12**

2026-10-28

**Changed**

* Server-side secrets now live in one encrypted store. This release is groundwork for encrypting environment secrets, which arrived in v0.0.13.
* More than one dashboard replica is supported for self-hosted installs that use S3 for artifacts.

**v0.0.11**

2026-10-27

**Improved**

* **Gates no longer run twice.** If the agent already passed a gate on the final files, completion reuses that result. See [Gates](https://docs.bellows.cloud/guides/gates.md#when-gates-run).
* Gate runs and pull request actions are recorded and shown in the task stats and activity.
* Claude Code can run gates and publish itself.

**Fixed**

* Claude Code's command allowlist refused the gate and publish commands.
* A runner taken over after a driver restart kept an old lease, so Jira and GitHub refused it.
* Transcripts are no longer lost when a task volume is cleaned up.

**v0.0.10**

2026-10-26

**New**

* **Gate filter.** A gate with `filter: true` lets the agent run part of a suite. The value arrives as `$GATE_FILTER`. A filtered run never counts as proof. See [Gates](https://docs.bellows.cloud/guides/gates.md#filters).
* The agent runs the gates that cover its change before it pushes.
* **Token accounting** by task, run, user, day, model and source.

**Improved**

* Tasks that changed nothing skip the end-of-run gates.
* Executor images ship `python3` and `procps`, and OpenCode ships `ripgrep`.

**v0.0.9**

2026-10-24

**Improved**

* **More resilient runs.** Setup, gates and services run in the task's own runner, with a checkpoint of each attempt. A lost node interrupts a task and resumes it instead of losing it.
* Limits on how many tasks run at once, in total, per organization and per repository.
* The runner's GitHub token renews during long runs.
* The agent's commands run under their own user.

**v0.0.7**

2026-10-14

**New**

* **On-demand gates.** Mark a gate `on_demand: true` and the agent runs it only when needed. See [Gates](https://docs.bellows.cloud/guides/gates.md#on-demand-gates).
* CLI: retry, reopen, answer questions and read prompts from a file or stdin.

**Changed**

* A follow-up on a task that still has an unfinished run is now refused with a `409` instead of queuing a second one.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.