Skip to content

What changed

Berd ships several releases a week. This page reads each release from the repository's CHANGELOG.md and says what you will notice, in plain words, with the build plumbing left out. Read it before updating, and read the matching Coach pages after.

Two reading notes. The date on a section is the changelog's own date, which can be a few days before the build reached the download page (v0.6.3 is dated 2026-09-05 and was published 2026-09-08). And the opening sentence of the v0.6.1 and v0.6.2 sections in the changelog names the previous version number; the bullets are right, the intro line is not.

v0.6.4 — 2026-09-10

What you will notice:

  • Artifact actions are back. Open in editor and the row actions on artifacts work again, including in a popped-out session window. They grey out when the file has been deleted on disk and return if it reappears.
  • An MCP App must ask before speaking as you. An app rendered inside the chat now needs your explicit confirmation, showing the exact text, before it can submit a message under your name.
  • Voice sounds like a person. Streamed replies are spoken in whole paragraphs instead of stray opening words; voice settings have Local and Cloud labels with Apple speech as the recommended default, and a reset that keeps your credentials.
  • Pi at onboarding. Pi can be installed with one click during onboarding, and the wider model picker makes long model names readable. GPT-6 Astra can be chosen for a Codex session.
  • Archived chats recover on send. Writing into an archived chat restores it to the active list. A remote session that no longer exists shows a read-only notice instead of a raw error.

Guide pages affected: Work with artifacts and MCP apps · Hold a voice conversation · Install Berd and get through onboarding · Choose a harness for a session · Keep sessions in order · An app in the chat wants to send a message as me

v0.6.3 — 2026-09-05

What you will notice:

  • Voice Conversation is real. Apple speech recognition and Siri voices, or OpenAI speech-to-text and text-to-speech; Berd starts speaking while the reply streams; a call keeps going when you switch chats or apps, with floating mute, hang-up and return controls. AirPods mute works on supported Macs.
  • Search covers everything. Cmd+K and Session History now search the full set of sessions, not only the ones already loaded. Recently used models sit near the top of the picker.
  • Agents travel and wear faces. Import an agent from a ZIP or by drag-and-drop; a custom gloopie is reusable across agents; a newly saved agent opens on its profile.
  • Home starts tidier. New installs get a refined starter layout, and you can place text labels on the canvas. Onboarding enters Berd directly after the welcome page.
  • Two new bundled skills and one published one. berd-monitor and berd-orchestrator let an agent run longer work outside the turn; buzz-handoff brings a Buzz thread into a private Berd conversation and can send back an approved reply.
  • Quieter fixes that matter. Messages sent while a chat is starting are no longer stranded; the artifact viewer follows the disk; the Skill Builder warns before discarding edits; About has merged into System settings; public builds no longer show Block-internal skill discovery.

Experiments added, all opt-in: remote SSH sessions, OpenAI Realtime voice, pull request tracking, chat on canvas, prompt pins.

Guide pages affected: Hold a voice conversation · Keep sessions in order · Build an agent of your own · Share, import and keep your agents · Make Home yours · Run long work outside the turn · Hand work between Berd and Buzz · Write a skill and decide where it lives · Keep Berd current and report problems · Should I turn experiments on?

v0.6.2 — 2026-08-18

The release that made the source public.

What you will notice:

  • Seven crafted agents. Agt. Builder, Choosey, Copycat, Pushback, Tinker and Wildcard join Berdy as bundled personas; new Home canvases feature Tinker and Wildcard, existing layouts are left alone.
  • Agents survive Slack. Download an agent card as a ZIP so its instructions and settings arrive intact; PNG and Markdown remain in the same menu.
  • Connections are grouped. Company-managed services in one group, local MCPs for Goose, Claude Code or Codex in another, with a shared search and an agent-guided setup flow.
  • Agent Work details stay put. Expanding earlier steps no longer jumps the conversation; long tool details stay compact and scrollable.

Guide pages affected: Cast the right built-in agent · Share, import and keep your agents · Manage connections and extensions · Berd's built-in agents

v0.6.1 — 2026-08-17

What you will notice:

  • Queued follow-ups send themselves. A follow-up goes when the agent is ready, even if you have navigated away, and waits for a new chat to finish starting.
  • Steer with an empty composer. While an agent is working, Enter with nothing typed steers the first queued message into the running turn.
  • Long chats stop losing their place. Tool-heavy transcripts keep accurate scrolling and visible history; an incomplete reload or compaction preserves what you have and shows a recoverable error.
  • Agent cards look the same everywhere. Shared and imported agents use one collectible-card design, show their real descriptions, and export to PNG dependably.
  • Installing and updating got smoother. A permanent download link for the latest macOS installer, and update-metadata fixes so older builds can take future updates.

Guide pages affected: Run a working session · I sent messages while the chat was starting · What does Steer do? · Install Berd and get through onboarding

v0.6.0 — 2026-08-14

What you will notice:

  • Worktree choice per folder. When creating a project with several Git folders, you choose how new chats treat each one, with clearer prompts and recovery when a chat starts.
  • Chats archive themselves. Settings → archive decides when inactive, unpinned chats are archived; active, running, pinned and draft-bearing chats are protected, and anything archived can be restored.
  • You can see who is doing what. Subagent activity names the subagent and its task, with accurate labels for delegation, waiting, interruptions and cancellations.
  • Share cards for everyone. Agent share cards no longer need an experiment.
  • A cleaner start. The sidebar opens straight into navigation and chats; a five-step Berdy tour introduces providers, chats, agents and skills, and guided starter tasks can be restored from the Home widget picker after dismissal (both experiments at the time).

Guide pages affected: Create a project around your folders · Keep sessions in order · Run long work outside the turn · Make Home yours

Outlook — interpretation, not a commitment

Read across the five releases, the direction is plain: voice as a first-class surface, agents as portable objects, one narrow bridge to Buzz, and more of the app steerable by the agent through berdctl. Several things ship first as experiments on public builds (chat on canvas, prompt pins, PR tracking, remote SSH) and some capabilities are compiled only into Block's internal builds (automations, in-app feedback, managed connections, skill discovery); which of those graduate to the public app is not stated anywhere. This paragraph is the Coach's reading of the release notes, not shipped features or commitments by Block. Last checked: Berd v0.6.4, 2026-09-11.

Last checked: Berd v0.6.4, 2026-09-11.

Reference

NEVER BESTUCK AGAINCLICK ME

Guide built with Berd, Claude and Codex · verified against Berd v0.6.4, 2026-09-11 · read from source, not yet tested in the app · Buzz Guide · About · not affiliated with Block