New
- Building blocks: Ask for raw sound-design material, such as drones, impacts, whooshes or sci-fi blips, and Reagent plays your own synths and effects with moving knobs on new tracks and records the results into your REAPER project. Each request starts with a round of very different variations: pick one to take it further, or say what to change, such as “darker” or “less reverb”. Learn more
- Plugin preset scan: Reagent scans the presets of your installed plugins in the background, on Mac and Windows, so building blocks start from sounds in your own library. Settings > Plugins shows each plugin’s progress and preset count, and lets you search, rescan or leave plugins out. Learn more
- Claude Sonnet 5.5: Reagent now runs on Claude Sonnet 5.5.
Improvements
- In Default permissions, setting up a building block on Reagent’s own new tracks runs without asking at every step. Recording still asks first.
- The plugin scan pauses while Reagent records a building block.
- When Reagent lists your installed plugins, each product appears once, and the list can be narrowed to instruments, effects or a name.
Bug Fixes
- When a script Reagent runs in REAPER hits an error, its changes stay in a single undo step and the arrange view refreshes.
- The REAPER bridge no longer stops working after a command fails, including when the SWS extension isn’t installed.
- Confirmation dialogs open in place instead of sliding in from the corner of the window.
New
- REAPER setup on every computer: Reagent walks you through setting up REAPER on each computer you use. It checks the REAPER folder, the SWS extension, the bridge and auto-start for you.
- Credits per reply: Each reply shows how many credits it used, with a running total while Reagent is still working.
Bug Fixes
- Starting the bridge while it’s already running restarts it instead of running a second copy that could make every change happen twice.
Everything here is for Enterprise accounts using a Custom Endpoint.
New
- Reasoning models: OpenAI reasoning models can now use the tools available on Enterprise, including Wwise questions, when your endpoint offers OpenAI’s Responses API. With API on Auto, Reagent picks the API for each model after testing your endpoint.
- More endpoint settings: Settings > Endpoint and the Custom Endpoint window add API, Reasoning effort, Max output tokens and Context window (the last two under Advanced). A Detected row in Settings > Endpoint shows which API Reagent uses and why.
Improvements
- Apply tests that your endpoint can handle Reagent’s tool requests before saving, and shows the endpoint’s own error if it can’t.
- Reagent retries a rejected request once in a format your endpoint accepts, and remembers what worked.
- The context usage indicator uses the context window you set for your endpoint, or a default for your model’s family.
Bug Fixes
- Fixed Claude responses getting cut short by a low default length limit on your endpoint.
- Fixed Gemini models failing partway through multi-step tasks.
- Fixed Wwise questions failing on endpoints that strictly check Reagent’s tools.
- Endpoint errors show the endpoint’s actual message, not unrelated advice about Reagent credits.
New
- Claude Sonnet 5: Reagent now runs on Claude Sonnet 5. The Wwise subagent and the Analyst moved to a newer, faster model.
- New look: A black-and-white design that matches the website, with redesigned sign-in, settings and onboarding screens and a new startup animation. Keyboard focus is clearer, the app follows your system’s reduce-motion setting, and on macOS you can zoom the whole interface from the keyboard.
- Redesigned sidebar: Chats are grouped by date and searchable, and you can star chats to pin them at the top.
- Collapsible activity steps: Several steps fold into one expandable summary, and steps that return tracks list them by name with their item counts.
- Sharper analysis: Loudness measurement is several times faster and more accurate on bass-heavy material. Transient detection catches quieter hits in takes that swing between loud and soft. Classifying many tracks returns a quicker, far cheaper summary.
- Browser sign-in: You sign in through your web browser, which also lets Enterprise accounts use their company’s single sign-on.
- Custom Endpoint: Enterprise accounts can now connect Reagent to their organization’s own model endpoint. Reagent sends its AI requests there instead of to Reagent’s servers, and doesn’t offer the features that send your content to Reagent’s servers: dialogue transcription, transcript analysis and translation, reference-file uploads, REAPER API search, the Wwise subagent’s online lookups and in-app feedback. Some services are still contacted, such as Reagent’s account service for sign-in and your profile, WorkOS for company SSO, reaperagent.com for release notes, GitHub for updates and PostHog for the analytics library’s settings (no usage events are sent). See where your data goes.
Improvements
- Fewer credits and faster responses on most tasks.
- REAPER edits and analysis run in your chat instead of being handed off, so tasks finish faster and more reliably.
- A failed step shows a short explanation and, where available, a suggested next step.
- Reagent explains file types it can’t attach, warns you before you send oversized attachments and lets you remove them without starting over.
- Long pasted text collapses into a block.
- When REAPER or Wwise isn’t connected, Reagent tells you what it can do and how to connect.
- Wwise scripts now run in an isolated sandbox for added safety.
Bug Fixes
- If part of a longer dialogue analysis fails, Reagent retries just that part and merges it back correctly.
- Asking for only complete or incomplete takes in the transcript no longer comes back empty.
- Comped takes on fixed lanes return the right items even when nothing is selected in REAPER.
- Several REAPER agent actions sent at once no longer interfere with each other.
- Chats that can’t be named automatically take their title from your first message, and generic titles get fixed when you reopen the chat.
- Pasted text keeps its line breaks.
- Permanently delete your account and billing yourself from your account page, which also cancels any active subscription.
- Add-on credit top-ups on every paid plan, not just the top tier.
New
- Context-aware live Wwise mixing: Reagent targets the exact sound variant you just heard in your game, whatever you have selected in Wwise.
Improvements
- Improved credit usage for the dialogue pipeline.
New
- Translation PDFs in non-Latin scripts: Dialogue Manager translation PDFs include fonts for Thai, Japanese, Chinese, Korean, Arabic, Hebrew, Hindi and other non-Latin scripts, and align right-to-left languages to the right.
- Tool permission modes: Choose how often Reagent asks before making changes: Default permissions, Auto-review or Full access.
- Cancel an agent question: Dismiss a mid-task question with Cancel or Esc, and the agent stops cleanly instead of guessing past you.
Improvements
- Faster responses.
- A clearer approval dialog that shows what Reagent is about to do.
- A smaller Windows installer that downloads and installs faster.
Bug Fixes
- Reference scripts (PDF, DOCX, TXT) attached to a chat now inform the dialogue pipeline, fixing inaccurate edits.
- Chat content no longer overflows narrow windows.
- Resume after approval works reliably across the whole turn.
- Chat session state persists across reloads.
- Cancelled questions no longer keep the timer running.
- Various Windows path handling fixes.
Everything here is for the dialogue editor, which isn’t available on Enterprise accounts.
Improvements
- Podcast section mapping better handles off-script content and preambles, and keeps the first word of scripted content.
Bug Fixes
- Fixed multi-mic podcast sessions where one speaker’s track could collapse into a single incomplete take.
- Dialogue cuts have 20ms edge fades to remove click and step artifacts at midpoint cuts.
- Track and item colors apply correctly after dialogue editing.
- Various dialogue pipeline reliability fixes.
- Slash Commands: Save your most-used prompts as reusable commands and trigger them with
/in chat. - File Operations: Reagent can move, copy, rename and create folders for you.
- Render media: Render selected items, tracks, regions, the time selection or the whole project to disk, with optional sample-rate, channel and dither overrides.
New
- Wwise integration: Use Reagent to debug issues, brainstorm Buss or Container structures, check implementation and duplicate objects with the same settings and properties. It works on Windows and macOS with Wwise 2021.1 or later.
Improvements
- Subagents work more efficiently and use fewer credits.
Bug Fixes
- Fixed tool names in chat.
New
- Translation for non-English audio: Reagent detects the language during transcription and offers an English translation, shown in a Dialogue Manager viewer with seek-to-line playback and PDF export.
- Multi-item podcast analysis: Podcast mode analyzes several items as one, for more accurate alignment and edit suggestions on recordings split across items.
Improvements
- Voice-takes mode rebuilds a reference script when none is provided, which greatly improves take classification.
- Reagent edits media items directly.
- The timer pauses while Reagent waits for your input.
- A desktop notification appears when Reagent asks a question and the app isn’t focused.
- Better dialogue pipeline results on longer recordings.
- On a track with multiple lanes, Reagent uses the comp lane by default.
- More credit usage improvements for the dialogue pipeline.
Bug Fixes
- Fixed rate limit messages showing as quota exhaustion.
New
- Podcast and narration suggested edits: Suggested cuts show as red (incomplete) items in REAPER, so you can preview them before committing.
Improvements
- Better dialogue pipeline accuracy, especially on longer media items.
- Dialogue analysis asks you to choose podcast or voice-takes mode, for more reliable results.
- Better voice-takes accuracy from cleaning up the script before analysis.
- Better dialogue pipeline results when no script is provided.
- The dialogue pipeline works on video media items.
Bug Fixes
- Fixed overlapping takes in the dialogue pipeline after processing.
- Fixed text-based reference files sometimes failing to load.
New
- REAPER subagent: Reagent could hand complex multi-step REAPER tasks to a REAPER subagent (removed September 16, 2026; REAPER work now runs in your chat). See Multi-step requests.
- Copy chat: Copy your whole conversation to the clipboard in one click.
- PDF and document uploads: Attach PDFs, Word files (.docx) and Excel spreadsheets (.xlsx) in your chat. (On Enterprise accounts, Word and Excel files can’t be attached.)
Improvements
- Your unsent message is kept when you switch sessions, until you quit Reagent.
- Better understanding of REAPER tools and when to use them.
- Longer timeouts for long-running operations.
- Lower memory use from more efficient project snapshot handling.
- Tasks take fewer credits.
New
- Dialogue pipeline upgrade: Dialogue editing can follow custom instructions for workflows like speaker detection or topic segmentation, such as “Edit VO and add marker where the speech differs from the script.”
- File attachments in chat: Attach reference files such as scripts, spreadsheets and PDFs for Reagent to use during processing. (On Enterprise accounts, Word and Excel files can’t be attached and reference files aren’t used for dialogue editing.)
- Better REAPER context detection: Reagent reads your selected items and tracks, cursor position and time selection each time you send a message.
- Audio classification for more formats: Sound classification works with common audio formats such as WAV, MP3, FLAC, OGG and AIFF, not just WAV. (On a Mac, classification needs Apple silicon and macOS 14 or later. See System Requirements.)
- Create Variation tool: Creating variations is much faster, and you can chain prompts together.
Improvements
- Faster runs that use fewer credits.
- Chat titles are generated instantly from your first message.
- A Copy button on your messages.
- Less filler and more direct action, which uses fewer credits.
- More consistent tool selection.
Bug Fixes
- Fixed chat messages rendering incorrectly.
- Fixed the splash screen not being draggable.
New
- Cloud-powered dialogue editing: Transcription and analysis run in the cloud for faster, more accurate results, with no local models to download.
- Incomplete take preservation: False starts, mid-word cutoffs and interrupted takes stay in the edit, colored red in REAPER.
Improvements
- Dialogue editing results appear in REAPER as each item finishes, not after the whole batch.
- Progress indicators show detailed stages during dialogue editing.
- The context usage indicator persists across sessions.
Bug Fixes
- Fixed media items not being renamed to their spoken text after dialogue editing.
- Fixed the app redirect after authentication.
New
- LUFS loudness targeting: Set the Take Volume Leveler to a LUFS target, such as -24 LUFS for broadcast, and Reagent measures and adjusts automatically.
- What’s New dialog: After an update, a dialog shows what changed.
- In-app feedback: Send bug reports, feature requests or general feedback from the app.
- Context usage indicator: A ring next to the send button shows how much of the conversation window you’ve used. It turns amber at 80% and red at 95%.
Improvements
- A progress ring in the title bar shows update download progress.
- Toast notifications have action buttons, such as Install Now.
- A redesigned chat input with a toolbar row below the text field.
- The Take Volume Leveler shows status messages while processing.
- Media items in large projects load in pages, for faster performance.
- A changelog link in Settings.
- Long sessions no longer throw authentication errors.
- Better transcription performance and stability.
- A turn timer on chat messages shows how long each response takes.
- Improved progress tracking during batch tool operations.
- Auto dialogue editor upgrades: Voice take splitting handles full recording sessions, even 2+ hours, by processing segments in parallel. A script reference file (CSV, TSV, TXT or XLSX) improves take classification, and edited items are renamed to match what was said.
New
- ReaScript authoring: Reagent can create, save and register custom Lua scripts as REAPER actions from the chat. It can run a script (a Run script block) or draft one for you to review first (a Create script block). You can assign a keyboard shortcut on first save, and ask Reagent to update or extend scripts in your Scripts directory.
Improvements
- When an app update changes the Bridge, Reagent updates the files and asks you to restart the Bridge script in REAPER.
- Script save names sync with the description.
New
- Voice take splitting: Reagent splits dialogue and speech recordings into takes using word-level timestamps, trims silence and edits directly in REAPER.
- Pro+ plan: More monthly credits than Pro, for heavy users. See current plans at reaperagent.com.
Improvements
- Your credit balance shows in your dashboard.
Bug Fixes
- Fixed an issue affecting Bridge connections on Windows.
- Take Volume Leveler: Levels the dynamics of dialogue or vocal tracks with take volume envelope automation. It’s non-destructive, so your audio stays untouched.
- Improved FX parameter handling for more reliable plugin control.
- Media item data includes more track information.
- Subscription tier changes sync to the app instantly.
- Improved markdown rendering in chat messages.
- Better handling of ambiguous requests, with guided context gathering.
New
- Subscriptions and credits: Subscription plans with credit-based usage, a live credit balance and a clear message when your quota is exhausted.
- UCS audio classification: Classify the sounds in your REAPER project with the Universal Category System, using a model built into Reagent.
- File attachments: Drag and drop files or folders into the chat, or use the file picker, to attach reference material.
Improvements
- Delete a chat, or clear all chats, from the sidebar.
- Window position and size are remembered between sessions.
- Improved onboarding (since replaced; see Quick Start).
Bug Fixes
- Fixed traffic light button positioning on macOS.
- Fixed skills not being recognized in some configurations.
- First release: Reagent, an AI assistant that controls REAPER from plain-English requests, such as arranging items, processing audio or managing tracks.
- Audio analysis: Sound type classification, peak and silence detection, transient detection, waveform extraction and automatic markers.
- Create variations: Randomized variations of selected media items, with adjustable pitch, volume and playback rate.