Skip to content

Releases: AhmadALSaffan/RAFIQ

Rafiq v0.5.14: answers from your files with their sources, and spending less

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 31 Oct 15:55

Rafiq v0.5.14: answers from your files with their sources, and spending less

New

  • Knowledge folder. Settings → Memory: choose a folder of PDFs, Word, PowerPoint, Excel and notes, and Rafiq answers from it. Every answer built on it lists its sources under the reply, with the file name, and the page for PDF and PowerPoint; a press shows the file. Reading and the index stay on your computer, and only the passages for your question go to the model. A scanned PDF with no text is listed as unread instead of silently missing.
  • What Rafiq knows about me. A page (Ctrl+K → "What Rafiq knows about me") with everything it remembers: edit any line in place, switch it off, or erase it.
  • Smart route. Settings → Usage: pick a cheap model, and short questions that need no files or tools go to it while real work stays with the model you chose. It's decided by plain rules, with no model call, and when in doubt it stays with the strong one. Every routed reply says who answered, with a press to ask the strong model instead. If the cheap model fails, the strong one takes over.
  • A spending cap per task or chat. Settings → Usage for all tasks and chats, /settings for one chat. When it's reached Rafiq stops and asks; if you allow it, it goes on until it has spent as much again. It's checked before each model request, so a long reply can pass it a little, and models with no known price don't count.
  • Compare two models. The same question side by side, with the answer, the price, the total time and the time to the first word. From the Models page. Nothing is saved.

Updating

Rafiq offers the update by itself. Or download Rafiq_0.5.14_x64-setup.exe below and check it:

Get-FileHash .\Rafiq_0.5.14_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.13: models that argue, a radio show about your day, and an assistant that notices

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 30 Oct 19:48

Rafiq v0.5.13: models that argue, a radio show about your day, and an assistant that notices

New

  • Model council. A new page in the sidebar. Ask a hard question and two or three of your models each answer on their own, then critique each other under neutral names so nobody defers to a famous model, and one writes a decision with the disagreements left in view. It ends as an ordinary chat you can carry on, with every answer kept above the decision. Each model is charged for it, so a question costs more than a normal chat.
  • Morning radio. Turn it on in Settings → General. Every morning at your time, a short spoken show about your day: what got done, what failed, what's on today, the files that changed, and news on your topics if a web search service is set up. It's written only from what's really there. It waits on the home page; "ask about it" opens it as a chat with the voice conversation already listening. Writing it costs one model request a day, and a paid voice only counts when you listen.
  • Rafiq notices. Off until you turn it on. Rafiq spots what you keep doing by hand: the same request on the same weekday or most days, or the same kind of file moved to the same folder, and offers once to do it, as a schedule, a folder trigger or a template. It's all counting on your computer: no model is called, nothing leaves it, and nothing happens without your yes.
  • ElevenLabs can listen too. Settings → General → Voice → "Who listens" picks who turns your speech into text, for the mic and the voice conversation. With ElevenLabs for both, the whole conversation runs on one key.

Fixed

  • The mic and the voice conversation failed with a Groq model: the recording went to OpenAI with the Groq key.

Updating

Rafiq offers the update by itself. Or download Rafiq_0.5.13_x64-setup.exe below and check it:

Get-FileHash .\Rafiq_0.5.13_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.12: sees your screen and talks with you

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 29 Oct 18:21

Rafiq v0.5.12: sees your screen and talks with you

New

  • Ask about your screen. A new shortcut (Ctrl+Alt+S, changeable in Settings → General) takes a picture of the screen under the mouse. Drag out the part you mean (Enter takes the whole screen) and the quick ask box opens with it attached, with one-press actions: explain it, what's wrong here, copy out the text, translate it. Needs a model that can see images.
  • Voice conversation. The headphones button next to the mic starts a hands-free conversation: Rafiq listens until you stop talking, answers out loud, and listens again. Every reply also gets a "Listen" button. Windows' own voices are used by default (free, nothing leaves your computer); when Windows has no voice for the language — common for Arabic — Rafiq uses the local voice from Video settings or an OpenAI voice. Settings → General → Voice picks who reads.
  • Tasks that start themselves. Schedules can now start on an event, not only a time: a new file in a folder (with optional patterns like *.pdf), copied text (any, English/Latin, Arabic, or a link), or a program opening. Files already there, text already copied and programs already open never start anything; downloads are waited for until they finish. Each trigger waits between runs and starts at most 30 tasks an hour. A trigger task's result shows in the Windows notification.
  • Find your files by what's in them. Ask "where's the March invoice?" and Rafiq finds it from what's written inside. It keeps an index of Documents, Desktop and Downloads (or the folders you choose) on your computer only, reading text, Markdown, code, CSV, Word, PowerPoint, Excel and PDF; other files are found by name. The model searches with every way the thing might be written, in Arabic and English. Settings → General → Search your files.
  • Up to 4 chats side by side on the chat page.
  • Richer 3D in Motion. Ready-made trees, bushes, rocks, mountains, clouds, houses, buildings (8 forms, 8 tops, 5 façades), whole cities, cars, people, street lamps, roads, fences, benches and water that reflects; boxes are rounded by default; and a full 3D guide. The model can look up a real photo (motion_reference) and copy a place tower by tower.

Fixed

  • Switching the theme could leave the globe blank until Rafiq was restarted.

Updating

Rafiq offers the update by itself. Or download Rafiq_0.5.12_x64-setup.exe below and check it:

Get-FileHash .\Rafiq_0.5.12_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.11: lighter on your computer, quicker at work

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 28 Oct 11:22

Rafiq v0.5.11: lighter on your computer, quicker at work

A release about performance. On a laptop with built-in graphics, Rafiq used to keep its
heavy parts (the motion engine, 3D, the globe behind the pages) busy even when you weren't
using them. Now it runs only what you have open.


⚡ Only what you open runs

  • Pages load when you open them: Motion, the creative studio, design and notes are no
    longer part of what loads at startup. The code loaded when the app opens went from
    2.8 MB to 1.3 MB.
  • The motion engine lives inside the project: it starts when you open a motion project
    (a short «Starting the motion engine…» while it does) and stops when you leave it. Outside
    a project, the model checks a reel's structure and asks you to open it to measure frames.
  • Motion cards show a still of the reel, saved the first time you open the project,
    instead of running a player each.
  • Design previews run only while they're on screen, in the chat and in the designs list.
  • Settings → Motion shows its preview only when you ask for it.

🌍 Quicker between pages

  • The globe behind the pages is built once: pages that cover it only hide it, so moving
    between pages no longer creates a new graphics context each time (and the old one is
    released right away).
  • It draws at 30 frames a second and one pixel per screen point: its graphics work went
    from 19% to 8% on the development machine.
  • After you switch pages it waits a moment, so the new page draws first.
  • Where the window has no graphics card (software drawing), it stands still.
  • It stops drawing while Rafiq is in the background.

🎬 Motion doesn't freeze

  • The engine gives the window room while it checks a reel, so the app stays responsive.
  • One engine stays up between the model's checks instead of starting over each time.
  • Every 3D preview shares one graphics context.

🛠 Quicker, cheaper tasks

  • Edit part of a file: a new tool replaces just the text that changes instead of
    rewriting the whole file.
  • Scripts instead of one-liners: the model sends Python, PowerShell or Node code as a
    script, which runs from a file with UTF-8 set up, so Arabic text no longer breaks it
    (no more UnicodeEncodeError or quoting errors from cmd).
  • Long tasks keep their context small: older tool results are condensed as a task goes
    on, so long tasks use fewer tokens.
  • A script that deletes files needs a press-and-hold, like a risky command.

🔧 Fixed

  • The glowing orb could stay stuck above a reply.
  • The What's new video lets go of its memory when the dialog closes.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.11_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.11_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.10: smarter design, and motion that shows what's going on

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 27 Oct 18:16

Rafiq v0.5.10: smarter design, and motion that shows what's going on

Sites stop looking alike, you talk to a design part by part, and the whole app now shows
what it's doing while it works. A one-minute video of everything is attached
(whats-new-en.mp4, also in Arabic and Russian).


🎨 Design

  • Every site gets its own character: each new design gets one of 13 art directions (a
    run of posters, a living bento, a split screen, a manifesto in words, a horizontal
    gallery…), picked so it differs from your recent designs. The model is told how to
    compose the page in it, not just which colours to use.
  • 3 variants: one button asks for three versions of the page, each in a different
    direction. They show side by side, any one can be enlarged, and «Keep this one» makes it
    the page (and its direction the design's).
  • Pick parts and say what to change: click several parts of a page or of an app's
    screens; each gets a number on the design and its own note in a small composer, and they
    all go to the model in one message with each part's selector and HTML. Pointing at a note
    scrolls to its part and pulses it.
  • Undo under every design reply: takes every document the reply changed back to its
    previous version; the notification has «Redo». Nothing is deleted: every version stays in
    the Versions tab.
  • Before & after: the old and new versions one over the other, with a handle to drag
    across them; both pages scroll together.
  • The page builds while it's written: instead of a blank wait, the preview shows each
    part (header, sections, footer) as it arrives, the one being written shimmering.
  • The preview frame springs between phone, tablet and desktop widths instead of jumping.

💬 Chat

  • What it's doing, at a glance: the status line's icon follows the kind of work
    (reading, writing, running, the web, waiting for you), with a count of finished steps.
  • Tool cards: a small pill while a step runs, a card with its result when it lands, and
    a result that went fine folds away once the next step starts (a failure stays open).
  • Risky commands take a long press: deleting files, killing a process, pushing to git,
    posting to a team channel, and commands like rm, del, Remove-Item,
    git push --force or git reset --hard are allowed with a press-and-hold, so a stray
    click allows nothing.
  • Context ring: beside the model, how full its window is after a reply (amber past 70%,
    red past 90%); a click summarizes the older messages to make room.
  • Your message flies from the box to its place in the chat, and lands once.
  • Done while you were away: a Windows notification and a flashing taskbar button when a
    reply or a design finishes while you're in another window.
  • When a file is written with the side panel closed, its name slides into the panel's button.

✨ Across the app

  • Notifications with an action (like «Redo») show how long they'll wait, and pause with it.
  • Copy buttons turn into a check that draws itself.
  • More switches slide between their options.

🔧 Fixed

  • A two-finger touchpad drag could pull the whole window aside and show black behind it.
  • On the phone board, scrolling with the mouse selected text; dragging now scrolls the screen
    (or moves the board where a screen doesn't scroll), and clicks still work.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.10_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.10_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.9: design your app on one board

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 25 Oct 22:08

Rafiq v0.5.9: design your app on one board

A new door in the creative hub: App design. Describe your app, and its screens come back
on one board, each in a phone frame, with the app's identity first.


📱 App design, on one board

  • Its own place: «App design» in the creative hub, with its own list and its own
    questions (what the app does, who uses it, the key screens, the feel, the language).
  • The identity first: the logo, the app icon, the colours by name, the type and the
    components, on a sheet of their own.
  • Every screen in a phone frame: in the order of the user's journey, numbered and
    named, with the island, the status bar and the home indicator drawn by Rafiq, so every
    frame matches.
  • Live screens: they scroll and move, and a button that leads to another screen takes
    the board there.
  • Move and zoom like a design canvas: drag to pan, Ctrl+wheel to zoom, one button for
    the whole board.
  • Export the board as one HTML file that opens in any browser, zoom and links included.
  • Models design the screens; they can't write files or run commands in an app design, and an
    answer that would break the board (a screen with no name, a phone drawn by hand, several
    screens in one, no identity) goes back to be redone.

✨ Also new in motion

  • New shapes: twelve more in 2D (diamond, bolt, drop, plus, moon, cloud, gear, squircle,
    burst, blob, speech bubble, shield) and 3D hearts, stars, bolts, gems and vases.
  • Music in ready styles: twelve of them (lo-fi, cinematic, epic, synthwave, house, trap,
    ambient, corporate, piano, Arabic, playful, tension), with Arabic instruments: oud, qanun,
    ney, darbuka, riq and daf.
  • Eight new sound effects: shimmer, sparkle, heartbeat, glitch, boom, swell, notify, shutter.

🔧 Fixed

  • A model could answer a motion request with HTML or three.js code instead of building the
    video; it now builds with the engine every time.
  • A 3D video exported at 120 fps stuttered on most laptops and 60 Hz screens; exports default to 60.
  • The timeline now follows the playhead while a video plays.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.9_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.9_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.8: 3D inside motion

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 24 Oct 21:15

Rafiq v0.5.8: 3D inside motion

Models now build 3D scenes in your videos, compose their own music, and work in a new
creative hub. The video at the top of the What's new dialog was made inside Rafiq.


🪐 3D inside motion

  • A new layer, scene3d: three.js renders it into the frame like a picture, in the
    preview and in the MP4 export alike, and only loads when a video has 3D in it.
  • Twelve ready looks: 3D logo, particle text, 3D type, galaxy, globe with flight arcs,
    solar system, block city, chrome blob, product spin, 3D cards, 3D bars, falling shapes —
    from a few params, with anything added on top.
  • Or from scratch: shapes, 3D text, extruded SVGs and icons, GLB models, particles,
    instances, tubes, globes; chrome, gold, glass, holographic and neon materials; lights,
    environments, camera moves, bloom and grain.
  • Real surfaces: a picture wrapped on a sphere (a planet's map), bump and night-light
    maps, planet rings. Models can bring a picture in from a link (motion_fetch_image,
    approved by you each time).
  • 3D Arabic type joins and reads right to left (shaped by HarfBuzz).
  • Upload a .glb and it comes in turning on a pedestal.
  • In the editor: a 3D section in the add panel, and the look, text, material, mood, camera
    move and glow in the layer's properties.

🎵 Sound, composed

  • compose: the model writes the score for each video: tempo, key, chords, melody,
    bassline, drums by the step — on 18 instruments it can shape and 14 drums.
  • No sound unless you ask: the guides no longer add music to every video, and effects
    can be retuned and stretched.

✨ Also new

  • The creative hub: designs, motion and notes behind one button in the rail, on a page
    with a clear way into each.
  • Your ChatGPT plan in sight: a ring beside a ChatGPT model shows the usage OpenAI
    reports, Rafiq's own requests and tokens, and a clear guide when you reach the limit.
  • Models by strength: strong models get the format and direct freely; lighter ones
    (DeepSeek, Qwen, Haiku, Copilot's Auto, small and local models) get step-by-step guides.
  • A new loader: an orb on the working line, rising out of the send button in motion.

🔧 Fixed

  • The timeline drew a late layer's keyframes past the video's end, so it seemed to go on
    forever; and nothing a model places can sit outside the video's time any more.
  • A ChatGPT limit reached mid-reply showed «ChatGPT returned an error: None».
  • Moving a clip on the timeline now carries its keyframes with it.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.8_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.8_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.7: ChatGPT models fixed

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 24 Oct 04:46

Rafiq v0.5.7: ChatGPT models fixed

A fix for Sign in with ChatGPT: a connected ChatGPT account now lists its models, and
adding one works.


🔧 Fixed

  • Adding a model on a ChatGPT account failed. The agent didn't accept ChatGPT as a
    provider when adding a model, so fetching the model list and saving were both refused
    (and the form showed "[object Object]"). Now it lists every model your plan offers;
    pick one, with nothing to type.
  • The model list stays fresh: the one-hour sign-in is renewed before the list is
    fetched.
  • Readable errors everywhere: an error the agent sends back as a list now reads as
    text, never "[object Object]".

✨ Smaller things

  • "Add model" on a connected account (Settings → Connected accounts) when no model uses
    it yet: it opens the model form on that account, with its models already listed.
  • Opening the model form on a sign-in provider (from the welcome too) picks your account
    and lists its models right away.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.7_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.7_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.6

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 23 Oct 13:22

Rafiq v0.5.6: Sign in with ChatGPT

Use your ChatGPT plan in Rafiq: no API key, and nothing extra to pay.


🔑 Sign in with ChatGPT

  • Connect a ChatGPT account in Settings → Connected accounts, or while adding a model
    ("Continue with ChatGPT"). You sign in on OpenAI's own page; Rafiq never sees your
    password.
  • Your plan pays for Rafiq's requests. Eligible requests count toward your ChatGPT plan
    usage, and you set Rafiq's limit in ChatGPT settings → Usage. It needs a plan that offers
    usage in other apps (Plus or Pro).
  • Pick any model your plan offers. It works in chats and tasks, with Rafiq's tools,
    pictures and reasoning levels.
  • It stays signed in: the one-hour access is renewed on its own, and disconnecting the
    account also ends the sign-in at OpenAI.

This uses OpenAI's documented "Sign in with ChatGPT" flow for open-source apps (OAuth 2.0 +
OpenID Connect with PKCE, and the public Responses API). Tokens are kept in the Windows
keychain like every other sign-in.

🎬 What's new

  • The What's new dialog can show a picture as well as a video. This one ships with the app,
    so it shows offline too.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.6_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.6_x64-setup.exe -Algorithm SHA256

Rafiq v0.5.5

Choose a tag to compare

@AhmadALSaffan AhmadALSaffan released this 21 Oct 21:58

Rafiq v0.5.5: Notes, in Arabic and English together

Most Markdown editors scramble a line that mixes Arabic and English: "C++" turns into
"++C", "8 MB" into "MB 8", and a sentence that starts with an English word reads backwards.
Rafiq now has notes of its own that get it right
(#1).


📝 Notes

  • A Notes page in the sidebar. Every note is a Markdown file in a folder of your own
    (Documents\Rafiq\notes by default, or any folder you pick, OneDrive included), so it
    opens anywhere and stays yours.
  • Every line takes its direction from its words, not from its first letter, and every
    English phrase inside an Arabic sentence (or Arabic inside English) stays in one piece,
    with its numbers, symbols and brackets on the right side. The same rules apply while you
    write and when you read.
  • Write or read (Ctrl+E). It saves as you type, and if the file changed outside Rafiq it
    asks which version to keep.
  • A formatting bar for headings, lists, tasks, tables, code, links and pictures. Paste or
    drop a picture and it's saved next to the note.
  • An outline of the headings, [[links]] between notes, and tasks you tick while reading.
  • Folders, search and pins. Drag a note onto a folder to move it. Search finds Arabic
    however it's spelled («اسامه» finds «أسامة»).
  • Export to Word or PDF.

✨ "Help me"

  • Select text and have it tidied, continued, translated, shortened or corrected, then
    choose whether to replace it or add the result after it.
  • Pick the model that helps, right in the menu. Rafiq remembers it, and the result says
    which model wrote it.

💬 Notes in your chats

  • "Save as note" on any reply.
  • @ has a "My notes" tab, in chats with or without a folder. The model reads the note
    you mention.
  • The model can keep notes for you. It writes a new note when you ask, and before it
    changes one you already have, it shows you the difference and waits for you.

🪟 Markdown files from Windows

  • "Open with → Rafiq" for .md files. Rafiq doesn't make itself the default program.
  • Rafiq Lite: one file in a light window of its own, just the editor. It's in the
    right-click menu of every Markdown file. When you open a file with Rafiq, it asks whether
    you want the light window or the full app, and can remember your answer (Settings →
    General).

🎬 What's new, after each update

  • The first launch after an update shows a short video of what changed, in your language,
    with the main points under it. It stays until you close it, and the About page opens it again.

The notes editor is built on CodeMirror (MIT). See
THIRD_PARTY_NOTICES.md.

Updating: installed copies offer this update on their own. For a new install, download
Rafiq_0.5.5_x64-setup.exe below and check it against SHA256SUMS.txt:

Get-FileHash .\Rafiq_0.5.5_x64-setup.exe -Algorithm SHA256