Download Rephlo Free - The AI Workspace for Windows, macOS & Linux
Download Rephlo free for Windows, macOS, and Linux. An AI workspace — chat, document-grounded Knowledge Spaces, and screen-aware Visual Intelligence — that works in any app. Version 1.7.1, released .
File size: ~227 MB. Format: .exe.
System Requirements
Windows - Windows 10/11 (64-bit)
- Windows 10 (version 1903 or later) or Windows 11
- 64-bit processor (x64 or ARM64)
- 4 GB RAM minimum (8 GB recommended)
- 500 MB free disk space
- Internet connection for BYOK providers
- Optional: GPU for local Ollama models
macOS - macOS 11.0+ (Big Sur or later)
- macOS 11.0 Big Sur or later
- Intel or Apple Silicon (M1/M2/M3)
- 4 GB RAM minimum (8 GB recommended)
- 500 MB free disk space
- Internet connection for BYOK providers
- Optional: Apple Silicon for optimized local AI
Linux - Ubuntu 20.04+ or equivalent
- Ubuntu 20.04 LTS or later (or equivalent distro)
- 64-bit processor (x64)
- 4 GB RAM minimum (8 GB recommended)
- 500 MB free disk space
- glibc 2.31 or later
- FUSE 2 for AppImage support
Trust and Security
- Code Signed: Digitally signed for Windows and macOS
- No Telemetry: Zero tracking, zero analytics, zero data collection
- GDPR Compliant: Full compliance with European data protection
- Local Encryption: API keys encrypted on your device
Changelog
v1.7.1 -
- Personal information is now removed before cloud requests by default — redaction used to sit off until you went looking for it; email addresses, phone numbers, government ID numbers and full card details are found and masked on your own computer before a request leaves it, and you can still turn it off in Settings ▸ Privacy
- Names, dates of birth and addresses can be masked too — these have no fixed shape, so they are detected by a separate on-device model you download once; until you do, those categories are shown as unavailable rather than quietly skipped
- Attached images are covered by redaction as well — Rephlo reads the text in an image on your machine first, and if it finds personal information the image is replaced by its masked text instead of being sent as it is
- Attach an image to a model that cannot see it — Rephlo now reads the text out of the image on your computer and passes that along, instead of dropping the image and letting the model answer confidently about something it never saw; when there is no readable text the model is told exactly that, so it says it could not see the image instead of inventing an answer
- On-device features answer sooner and sit lighter — the local models behind redaction and document search begin loading while you type instead of when you press send, and they hand back their memory after a stretch of inactivity
- Switching an on-device feature off now frees its memory straight away, rather than leaving the model loaded for the rest of the session
- A smoother chat — the divider under each assistant reply is gone, so a conversation reads as one continuous flow
v1.7.0 -
- Rephlo can now read your scanned documents — a PDF that came from a scanner, a photo, or an export with no text layer is no longer just a picture; you can search it, ask questions about it, and use it as context in a Knowledge Space
- The reading happens on your own computer — scanned documents are never uploaded to be recognised; the text is extracted on your machine and stays there
- Turn it on or off in Settings ▸ Privacy — look for "Read scanned documents on this device"; it is on by default, and with it off scanned files are sent to the AI model as before
- The first time you use it, Rephlo downloads a small text-recognition model — about 15 MB, once, in the background, and it tells you when it is ready
- Scanning in another language — "Scanned document language" follows your app language by default, and you can pick a different one if you scan documents written in it
- Very long documents are marked, not silently cut short — if a scan runs past the on-device page limit, Rephlo indexes the text it already has and shows a "Partial OCR" badge in the Space
- The template gallery is now fully translated into Korean, and a number of gaps in the German translations have been filled in
- Steadier attachments and Spaces — several cases where a document could be indexed incompletely, or counted twice, have been corrected, and cancelling a save no longer leaves a Space half-changed
- Exported files no longer contain folder paths from your computer, image size choices are remembered, and the check for updates on macOS and Linux no longer reports problems that were not real
v1.6.1 -
- A clearer image picker — the model you are about to use is now marked with a checkmark and a highlight, and the chip beside it shows the size you picked (and the quality, for models that offer it)
- The Create image menu stays put while you choose — picking a model no longer closes the menu early; it stays open until you click Done, click away, or press Esc
- Image tools in your language — the size and quality options, their tooltips, and the messages shown while an image is being created are now translated in all nine languages
- The "Generating image…" tile stays inside its own frame — the loading bar no longer streaks past the edges of the preview, most noticeable on tall and wide shapes
- Clearer messages when an image cannot be created — Rephlo explains it in a plain sentence instead of occasionally showing raw technical output
- A declined image is no longer reported as an outage — when an image is turned down by the content policy, Rephlo says so instead of claiming the service is temporarily unavailable
- Ordinary PDFs are no longer turned away as too large — Rephlo now measures a PDF’s actual pages instead of guessing from file size, so everyday documents like a 10-page scan are accepted
v1.6.0 -
- Rephlo now speaks nine languages — English, Vietnamese, French, Spanish, German, Japanese, Portuguese (Brazil), Simplified Chinese, and Korean, with six added in this release; Rephlo picks up your system language on first launch, and you can change it any time in Settings ▸ General
- Translated throughout, not partially — menus, settings, dialogs, buttons and error messages are all covered, and help, Buy Credits and Upgrade pages open in your language instead of dropping you onto an English page
- Create images right in chat — pick "Create image" from the + menu, describe what you want, and Rephlo generates it alongside your conversation; generated images use credits like any other request, and your plan sets the maximum resolution
- High quality on Pro and above — Pro, Pro+, Elite, Team, and Perpetual plans can select High image quality for sharper, more detailed results
- Edit an image by describing the change — attach a reference image and say what to change, instead of starting a new prompt from scratch
- Keep refining without starting over — follow up on a generated image across multiple turns and Rephlo builds on the previous version
- Clear content-policy messages — if a prompt is blocked, Rephlo explains it in plain language, in your own language, instead of showing a raw error
- Simpler default shortcuts — screen capture is now Ctrl+Shift+S (⇧⌘S on Mac) and the context menu is Ctrl+Shift+Space (⇧⌘Space on Mac); existing installations keep the shortcuts you already have
- Settings ▸ About now states plainly that Rephlo is an independent product, not affiliated with the AI providers it connects to
v1.5.1 -
- Fixed a crash when using connectors with Claude models — turning on a connector (for example Notion) while chatting with a Claude model could interrupt the reply with an error; chat now handles these responses smoothly
- Cleaner, consistent model ordering — the model picker now follows a single, predictable order across all models instead of only sorting within each group
- Check for Updates now takes Microsoft Store users straight to the Store to grab the latest version
v1.5.0 -
- Connectors are here — link Rephlo to 15 of your favorite tools (Notion, Linear, Stripe, Airtable, Todoist, Miro, Sentry, Canva, Supabase, Wix, Cloudflare Docs, Context7, Exa, Firecrawl) and let chat search, look up, and act in them, with one-click sign-in and per-conversation control
- Privacy, stated plainly — connector sign-in tokens stay encrypted on your device and are never sent to Rephlo; only tool names and short descriptions are shared so chat knows what your connectors can do
- Chat drafts now survive anything — saved as you type, restored after restarts, kept with the right conversation, and the text you just sent no longer reappears in the input box
- Report AI content in one click — a new "Report inappropriate content" link in Settings ▸ About and the profile menu's new Help submenu
- No more idle GPU usage — hidden animations now stop when Rephlo is minimized or out of sight
- Cleaner trial handling — the trial badge disappears when you sign in with a Pro subscription and returns correctly when you sign out
v1.4.2 -
- On-device models handle much longer conversations and documents — they now use a far larger context window by default, and you can set the size yourself in Settings ▸ On-Device Models
- If an on-device model runs out of room, Rephlo now tells you exactly how to fix it instead of failing with a confusing error
- Fixed a crash when continuing certain chats — running a text command or a Space-backed command and then continuing the conversation no longer stops with an error
- Visual polish — clearer contrast on pressed buttons and tidier gallery and tool dialogs
v1.4.1 -
- On-device models can now read your attached documents — attach a PDF or document in chat with an on-device model and it actually reads it, instead of quietly ignoring it
- On-device responses keep generating when you switch conversations mid-answer, and some on-device models use noticeably less memory while responding
- Model selection fixes: sending during a model switch now waits for the switch to finish, new chats keep the model you explicitly picked, and opening Chat reliably resumes the model you last used
- "Hide online models from Rephlo Gateway" now fully applies — if the model you were using is one the toggle hides, Rephlo switches you off it instead of quietly continuing to use it
v1.4.0 -
- Start your Pro trial right in the app — the welcome screen activates your free 14-day trial instantly, no sign-in or browser needed, and it even works offline
- Rephlo remembers the model you last used, and the clearer model picker shows which models your plan includes — it will never quietly fall back to one it doesn't
- Commands that draw on your Spaces now include more relevant context from your documents by default, for richer, better-grounded results
- Smoother onboarding — reliable first launch on a fresh install, "Already have a license?" opens the right settings page, and the chat setup prompt steps aside once you pick a model
- Linux polish — the applications-menu entry now launches reliably with the real Rephlo icon, and Rephlo explains keyring unlock prompts before they appear
- Fixes: a macOS first-launch crash, a Windows first-launch display glitch, steadier streaming with mid-answer cancel and consistent retries, and "Copy as Markdown" working on the very first copy
v1.3.0 -
- Rephlo is now available on Linux — the same experience Windows and macOS users have (global hotkey, screenshot capture, in-place text replacement, and system tray), and it keeps itself up to date automatically. Grab the Linux build from the downloads page
- Smoother tables while the AI is typing — tables render cleanly as a response streams in, with no columns jumping or getting cut off, and wide tables scroll sideways
- More reliable responses — if a reply is interrupted or a model hiccups mid-answer, Rephlo now tells you clearly instead of ending with a silent empty message
- Fixes: steadier long conversations that draw on large Spaces, better-timed Space suggestions that never block your send, and a fixed rare crash at a fractional usage limit
v1.2.0 -
- Connectors now ask before making changes — the assistant can read from a connected app (like Notion) to help you, but always asks for your approval before creating, editing, or deleting anything (allow once, allow for the session, or decline)
- Large Spaces are handled gracefully — if a command would send more of a Space than the AI model can take in at once, Rephlo lets you refine your request instead of failing, and splits large lookups so you still get a complete answer. Space size limits have also been raised substantially
- Simpler Space data mode with a new Smart Mode Advisor that recommends the best mode for your content, plus a heads-up before any action that would use credits (like compacting a Space)
- The indicator showing which Space context shaped a reply is now consistent across all chat views, with a tooltip explaining how Rephlo chose what to draw on
- Built-in Rewrite and Summarize now use your selected Space, just like your own commands
- Fixes: no more duplicate Space uploads, clearer import errors, more reliable connector sign-in, wider chat tables, readable code blocks in light mode, and clearer messages when your own-key model hits its size limit
v1.1.2 -
- Automatic updates: Rephlo now downloads new versions in the background and shows a "Restart to update" button in the title bar — one click to install. Turn it off anytime in Settings → General
- If an update downloaded while you were away, the "Restart to update" button reappears the next time you open the app
- Settings → About now shows the full version number (e.g. 1.1.2), so it is easy to confirm exactly which build you are on
v1.1.1 -
- Space context now follows your conversation — start a chat from a command with a Space attached and Rephlo keeps applying that context on every turn, with a pill showing what is in play
- Leaner Space handling: commands capture just the content they actually used instead of copying whole files — same results, less space on your device
- Fixes: hosted rewrites now match your own-key results, cleaner chat retries, steadier clipboard and international (Unicode) typing on Windows, and inline-code formatting that lines up correctly
v1.1.0 -
- Drag & drop files and entire folders onto any Space — Rephlo scans folders, shows you what will be added, and ingests it all in one step
- The model selector now shows a burn-ratio badge, so you can balance capability against credit spend at a glance
- Smarter streaming: if an AI response is cut off mid-generation, Rephlo detects it, surfaces a clear retry, and re-syncs your credits so you are not charged for what you did not receive
- Many chat, model-selection, Spaces, and Windows tray-icon fixes for a smoother day-to-day
v1.0.1 GA -
- Rephlo 1.0 is officially here — our first general-availability release, signed and ready for Windows and macOS (Linux coming soon)
- Knowledge Spaces got smart: drop in your documents and Rephlo now pulls just the most relevant passages into each prompt — indexed and searched entirely on your own device, so your content never leaves your machine to be understood
- Real-time sync: create or edit a command on the web and it shows up on your desktop in seconds — your library, collections, and Spaces stay in step everywhere
- Chat with your results: continue any transformation as a back-and-forth conversation that remembers the whole thread, images and screenshots included
- Ask about anything on screen: capture a region with a crisp freeze-frame and send it — text, images, or diagrams — straight to the AI
- 9 AI providers, your keys, your call — OpenAI, Anthropic, Google, Azure OpenAI, Groq, xAI, OpenRouter, a dedicated API option, and fully-offline local models via Ollama
- Privacy by design: API keys encrypted on-device, on-device document indexing, and a fully-offline mode where your content is never transmitted
- Faster and smoother throughout — quicker model loading, snappier command lists, richer Markdown in chat, and refined light & dark themes
v0.9.0-rc -
- Brand-new visual identity rolling out across the landing, pricing, and marketing pages
- Smoother first-time experience and refined onboarding flow
- Performance and stability improvements throughout the desktop app
v0.8.0 -
- Sign in once on desktop or web — your session follows you
- Cross-device library sync: your commands, collections, and Spaces stay in step
- Faster, smoother streaming responses for long-form prompts
v0.7.0 -
- 8 AI providers wired up: OpenAI, Anthropic, Google, Azure OpenAI, Groq, xAI, OpenRouter, and Ollama
- Knowledge Spaces — upload documents to ground every prompt in your own context
- Template Gallery — 500+ commands organized by category
v0.6.0 -
- AI Command Builder for custom prompt automations
- Universal desktop hotkey transformation on Windows and macOS
- Local Ollama integration for fully-offline workflows