Whisper Studio 2.4.1

A small release that stops document revisions from multiplying files.

• One deliverable, one file. Asking for changes to a document the assistant just saved — another layout pass, one more round of feedback — now updates the same file in place instead of leaving a folder full of v2, final, and clean copies. Every revision replaces the previous version at the same path, even when the assistant takes several attempts to get there. You only get a second file if you ask for a separate copy.
• Superseded drafts now count as leftovers in the assistant's cleanup pass, so a stray intermediate version gets removed instead of forgotten.

Details in #82.

Whisper Studio 2.4.0

A release built around documents: the assistant now shapes Word, Excel and PowerPoint files the way you describe them, and can change the ones you already have.

• Ask for the formatting you actually want. Cell shading, merged cells, borders, column widths, shapes, images, charts and number formats are all reachable now. Describe what you want in the file and the assistant builds it.
• Edit documents you already have. Point at a .docx, .xlsx or .pptx and ask for a change. The assistant reads the file's structure first so it changes the right cells, works on a copy so your original is untouched, and shows you the code on the approval card before anything runs. These scripts get a permission category of their own, apart from file writes and shell commands.
• Word documents come out richer. Table cells carry color, and headings, nested lists, header cells, quotes, inline code and captions all come through from a plain description. Word files now build entirely in Python, so they no longer depend on anything macOS-specific.
• Diffs open instantly, with a 5000-line file going from 208ms to 0.2ms, and show a clear message and a working Retry if a request ever does fail.
• Requests hold their connection reliably, clearing the intermittent "Load failed" that could reach any part of the app.
• Approval cards appear right below the discussion that produced them, so the Approve button is on screen the moment it arrives.
• Thinking summaries are back on the current Claude models, and the workspace card now names the tool, your connected folder and the path it could not use, with an option to stay put and retry.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v2.3.2...v2.4.0

Whisper Studio 2.3.2

A maintenance release focused on the workspace, approvals and workflow runs.

New

• Ultracode on Sonnet 5. Capability and reasoning level are now separate settings, so ultracode runs at each model's own top level. Existing installs pick this up with no config change.
• Your effort level applies everywhere, including every agent, subagent and workflow child agent, adjusted per model where a model supports a different range.
• rm always asks first. Deleting files now requires an explicit approval that no permission mode or session setting can bypass, covering rmdir, unlink, shred, wrapped and piped forms, and find -delete.
• Folder access asks once, then remembers. A grant covers everything beneath the folder you chose, and never a parent or sibling.
• Workflows can target a specific folder. Name a path in the prompt and the run pins to it, shown on the approval card and used by every agent it spawns.
• A notice when a turn is downgraded, naming the model and effort actually used when a budget fallback changes them.

Fixed

• The workspace panel stays in sync with the app and recovers on its own after a restart or a brief backend interruption.
• Approving folder access now creates and connects the folder, so the workspace opens as the card described.
• The conversation keeps its scroll position when you open or close the Tasks pane.
• A workflow you stop now reports as stopped.
• Agents running in parallel each keep their own working directory, so one agent's cd no longer affects another.
• Workflow costs now include cache charges.
• A model defined in a workspace's own settings uses the right provider.
• A workflow model override that cannot be used says so on the run's event stream instead of falling through.
• Data retention gives a clear message when AWS credentials are unavailable.
• The context report no longer counts recalled memory toward prompt duplication.
• Renamed model entries keep their top reasoning level.
• Ultracode is available on Sonnet 5 straight away on existing installs.
• Opening a folder is now correctly treated as a write action, so read-only agents cannot redirect a session.
• When the GitHub tools are unavailable, the assistant is told how to enable them rather than retrying the shell.

Improved

• Workflows run a full 16 agents at once. The scheduler and the underlying thread pool now share one limit, and the shared client paces itself against account quota rather than failing agents.
• The workspace disconnect control is clearer about what it does, and the collapse chevron is named as the way to just hide the tree.
• Connecting a workspace stores the path exactly as the server resolves it, so containment, indexing and language-server checks all agree.
• Continuous integration now covers the macOS-only paths, including document creation and file-manager integration.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v2.3.0...v2.3.2

Whisper Studio 2.3.0

Whisper Studio can draw

Some answers were never meant to be paragraphs. An architecture you are trying to picture, a trend you are trying to feel, a flow with four branches and a loop: prose describes these, but it makes you assemble them in your head one sentence at a time.

This release gives the assistant a pen. Ask how the pieces fit and a diagram appears in the conversation. Ask what the numbers did and a chart appears, drawn from the actual data rather than described around it. The picture arrives in the reply itself, inline, where the sentence explaining it would have been.

Two tools sit behind it:

• create_visual draws diagrams as hand-authored SVG: architecture and flow, comparisons, timelines, state machines, sequence and pipeline. Every shape and label is placed deliberately, so a box sits where it means something rather than where a layout engine dropped it. Nothing names a colour. The drawing borrows the palette from your theme, so it reads correctly in all eight of them, and the text inside stays real text you can select, search and copy.
• create_chart plots real data from a Vega-Lite spec, including figures read straight out of a spreadsheet you attached or computed on the spot with run_python. Colours, fonts, grid and legend styling come from the app theme automatically, so the chart is asked for in terms of what the data means, never what it should look like.

Every picture can be downloaded as SVG or PNG, or copied to the clipboard. Both tools run entirely on your machine with nothing fetched at render time, and neither stops to ask permission, because nothing is written and nothing runs on your computer.

New features

• Side-by-side diffs from the Git panel. Clicking a file's status letter in Git Changes opens a read-only diff against HEAD, the way VS Code shows one. The window is resizable and movable, remembers where you put it, and switches between side-by-side and unified. Long runs of unchanged lines collapse to an expandable marker.
• Attached files are read in full. Dropping a spreadsheet into the composer used to keep a header plus 20 sample rows and throw the original away, so a question about a total got a confident sum of the sample. On a 10,000-row sheet that answer was wrong by 99.8%. Uploads now keep their bytes, and any count, total, average or chart is computed from the real file.
• Workflow runs are visible. A launched workflow now appears in the Background Tasks panel, as an inline card in the chat, and in a run list under Settings. When one finishes it reports itself back to the assistant on the next turn, so a swarm you left running is something you get told about rather than something you have to go looking for.

Bug fixes

• Ultracode never actually launched a workflow. The workflow tools were missing from every request, so the model quietly substituted a single agent while its own instructions told it to orchestrate a fleet. They are on the wire now.
• Approving a card could leave the turn dead. A failed approval request escaped its handler, so the assistant was never told you approved: no message, no error, and every button on the next card stuck on "Running…" until a reload. The resume path now mirrors a normal send, so history, settings and recovery after a restart behave identically on both.
• The token counter was wrong. The composer read "4 in" for prompts of several thousand tokens, because it counted only the part that missed the prompt cache and ignored everything served from it. Cost was always calculated correctly. Tokens, cost and remaining context now live in one readout in the composer instead of two places disagreeing.
• A small on-device model answered "hey" with a JSON blob. Anything under 7B was installed with the full tool pool, which such a model cannot drive: it emitted the tool call as raw text into the reply. Models that small are now installed as chat-only, flagged in Discover before you download and in the model picker when you choose one. Models installed before this change are corrected once on launch.
• Documentation described things that do not exist. The approval flow was written up across four diagrams and the README as issuing a "nonce", which appears nowhere in the code, and several pages claimed chat always routes to Claude on Bedrock when it also serves GPT-5.x and on-device models. Both corrected, plus a new tutorial on triggering an Ultracode swarm.
• For contributors: eight modules resolved the sessions database path at import time, so tests could write into the real database of whichever checkout ran the suite. Tests are isolated from it now.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v2.2.0...v2.3.0

Whisper Studio 2.2.0

• Workflows now run within a token budget (600,000 by default) instead of unbounded, so a runaway script can't spend forever. Permission modes gained a dedicated workflow category: bypass launches with no approval card, auto only cards runs over budget, and dontAsk declines automatically.
• Approving a workflow can trust it right from the approval card with one checkbox, instead of a trip to Settings.
• Full documentation refresh: every page checked against the current app and corrected, three new tutorials (saving files and creating documents, branches and worktrees for parallel sessions, exporting/importing sessions) each with an interactive diagram, and updated screenshots throughout.
• Installing the Mac app is now documented as the recommended path, with a direct link to the Releases page; running from source with setup.sh remains fully supported.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v2.1.0...v2.2.0

Whisper Studio 2.1.0

• Saving a file just works now. Say where ("save it in Downloads") and it goes exactly there; say nothing and it lands in your Documents folder with the path stated in the reply. No more "connect a workspace" or "where should I save this?" questions for a simple file save.
• Documents are fully built before you approve. The approval card shows the exact destination and approving simply moves the finished file into place, so nothing can fail after you click Yes, and declining leaves your folders untouched.
• New tools produce real Word, PowerPoint, Excel, and PDF files directly, with a standard professional layout: A4 pages and Calibri 11pt text for Word and PDF, widescreen 16:9 slides, bold header row and fitted columns in spreadsheets. No more shell workarounds or leftover temp files.
• Every file the assistant creates is a clickable link that opens in its default Mac app.
• Export or import many sessions at once (zip backups), with selection unified in one toolbar.
• Your draft message no longer disappears when connecting or disconnecting a workspace.
• Settings panels have consistent spacing everywhere, and long skill or feature-flag descriptions collapse with See more.
• Scheduled jobs and subagents now load only the tools they need each turn, cutting their per-turn cost the same way chat already does.
• Fix: the workspace prompt now genuinely pauses and continues from where it stopped once you pick a folder.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v2.0.0...v2.1.0

Whisper Studio 2.0.0

• Cron and subagents now run on the same engine as chat. Both used to run their own separate, more limited execution loops; unifying them also means subagents and scheduled jobs can use local on-device models where they couldn't before.
• Heads up: scheduled (cron) jobs can now auto-execute actions that previously just failed, like file writes and shell commands, the same way subagents already do. Destructive actions (GitHub mutations, etc.) still always require a human, unchanged.
• Custom subagent types, written as simple markdown files, or defined on the fly by the assistant itself when nothing built-in fits, and promotable to a permanent type with one ask.
• One-click import from Claude Code or Codex CLI projects, bringing over instructions, skills, and MCP server config.
• Sessions are now exportable/importable as portable .jsonl backups.
• Finer-grained permissions: per-category approval defaults (e.g. auto-approve reads, always confirm deletes), plus a "test this rule" dry-run before saving and an optional justification note on each rule.
• Auto mode gets a circuit breaker. It now sees real turn context instead of judging each tool call in isolation, and stops asking for confirmation on repeat denials instead of grinding through them.
• MCP servers can now be remote (not just local processes), can ask a follow-up question mid-call, and support per-server/per-tool approval rules.
• Optional sandbox network policy to restrict which domains sandboxed commands can reach. Off by default; nothing changes unless you turn it on.
• Fix: local models that used to go silent after a tool result now get one automatic retry before giving up.
• Fix: subagents can now actually see files you attached to the conversation.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v1.3.0...v2.0.0

Whisper Studio v1.3.0

• Sessions are now @-mentionable. Type @ in the composer to pull another session's recent conversation in as context — filtered as you type, same as skills.
• Fix: a leading @word no longer gets misread as an attempt to force a nonexistent skill.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v1.2.2...v1.3.0

Whisper Studio v1.2.2

Feature: sessions can message each other

Any chat session can now send a short text message directly to another one. The model calls list_sessions to find a target, then send_session_message to deliver it — it lands in the target session's history and appears live if that session's tab is open, rendered as a distinct message card with the sender named and a link back to that session.

Fix: composer opened too tall

The chat input's auto-grow measurement could race the stylesheet loading on startup, occasionally computing a much larger height than intended and getting stuck there until you started typing. The composer now always opens at its proper resting height.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v1.2.1...v1.2.2

Whisper Studio 1.2.1

Fix: memory-fit warnings were far too conservative

The memory-fit grading added in 1.2.0 was treating every companion model (both speech-recognition backends, plus all four workspace-indexing models) as if they all stayed loaded in memory at once alongside a chat model. In practice only one speech-recognition backend is ever active, and the indexing models unload right after each indexing pass — so the tool was reserving far more memory than actually gets used, and it graded quants as "won't fit" that ran fine, including already-installed models that were clearly working.

Now only the currently active speech-recognition backend is counted. Quant and model rows also state plainly how much free memory each one needs ("Needs about ~9.1 GB free to run") instead of only a relative verdict.

Install

Download the DMG below and drag Whisper Studio to Applications. On first launch, right-click the app and choose Open (the build is not notarized).

Apple Silicon, macOS 14 or later.

Full changelog: v1.2.0...v1.2.1