New Computer Use: Flowly can now act on the Mac directly open apps, click buttons, switch windows, read on-screen text. The bot reads the same semantic UI structure as VoiceOver so it can act on labels you'd say out loud ('the Sign In button in Safari') instead of guessing pixel coordinates. Sandboxed by default; every action visible in the Activity tab.
New Live tool-turn panel: Every assistant message ships a collapsible panel that streams tool calls (file edits, shell, search, computer-use) in real time, with per-tool renderers. The Activity tab's audit view uses the same renderers so history looks identical to live chat.
New In-app file preview: Click a file path the bot mentioned (markdown link or inline code) to open a resizable, syntax-highlighted side panel. Relative paths auto-resolve against common project roots; directories still reveal in Finder.
New Artifacts cloud sync: Canvases, slides, docs, and code sync automatically across signed-in devices. Offline-aware queue, S3-backed previews, 10 MB size cap (up from 500 KB).
New Screen-aware Coach (macOS, opt-in): When you enable it, a native ScreenCaptureKit sidecar shares lossless captures of the window under your cursor with the model alongside audio, so tips can reference what you're actually looking at. ElevenLabs Scribe v2 STT adds [YOU]: / [OTHER]: speaker labels. Dual-Cmd forces a tip mid-session. Notch escalates to always-on-top with a 3-second watchdog.
New Knowledge Graph overhaul: Floating detail panel, click-through entity relationships, in-app entity deletion (cascades to triples), liquid-glass zoom controls, flicker-free node drag.
New What's new modal: First launch after auto-update opens release highlights with a hero matching the website's auth screen and a link to the full changelog.
New Welcome-screen otter mascot (opt-in): Vector otter toggleable from Settings Appearance. Eyes track your cursor while you type, mouth opens pink when you paste, blinks idly.
New Hey Flowly wake word: Production-ready on-device wake-word integration. Talk hands-off when the notch is live without holding Fn.
New Skill library expanded: 30+ new built-in skills covering finance modelling (3-statement-model, DCF, Excel authoring), GitHub workflow (PR review, repo management, code review), design and docs (concept diagrams, Excalidraw, PowerPoint authoring, nano-pdf), and SaaS integrations (Notion, Linear, Airtable, Google Workspace). Browse the full catalogue in Dashboard Skills.
Improvements Sandbox by default: macOS uses sandbox-exec, Linux uses bubblewrap; deny-list profile blocks writes outside the workspace. Master switch in Dashboard Settings. Skill marketplace surfaces per-plugin risk labels.
Improvements Composer attachments: Upload-first flow with per-file status (queued/uploading/ready/failed), 50 MB video uploads, correct serverId/conversationId on every upload.
Improvements Windows / Linux parity: Platform-aware default hotkeys, modifier labels, voice shortcut hints, and smart-pause process detection. Dedicated Windows tray icon.
Improvements Coach quota visibility: Live progress bar in Settings Usage and the sidebar usage popover. Pre-flight gating stops the session cleanly on STT 429.
Improvements Coaching settings is now a floating panel matching the rest of the app's side-panel pattern.
Improvements Dashboard regrouped: AI Tools and Security promoted to top-level groups. Pre-Sonoma macOS gets a friendly capability warning.
Improvements Shared AlertDialog adopts the liquid-glass theme app-wide with softer corners.
Fix Activity tab now scrolls instead of clipping; Radix ScrollArea wrapper repaired.
Fix Tool output rendering is cleaner: stray wrapper markup removed, escape sequences normalised, and excessively long blobs collapse to a short summary by default.
Fix Artifacts: 8 small UX bugs (delete confirm cancel, sort persistence, distinct empty/loading/error states, a11y labels).
Fix Coach screenshots use a quality + cursor-display target that keeps small on-screen text legible to the model.
Fix Granting macOS Accessibility from onboarding registers the global hotkey live; no app restart needed.
Fix Coach Start gated on Flowly AI gateway availability so it doesn't 404 immediately.
Fix Computer Use finds targets more reliably: matches against multiple element attributes (not just the visible label), suggests the closest candidates when an exact match misses, and avoids re-using stale element IDs after a window changes.
Fix Terminal emulators (iTerm2, Terminal, Alacritty, Kitty, Ghostty, Wezterm) get clipboard paste instead of synthetic key events.
-
Internal Sentry on artifact sync with scope tags. Persistent sync cursor avoids re-uploads after cold restart. Window-hidden poll pause for limits + bot updates. package.json at 1.4.0.
Flowly
To answer your self-host question: full offline capability with a local model, plus a visible action log I can actually audit after the fact - not just sandboxing. Right now the biggest trust gap with agents that touch my real data isn't whether they're capable, it's that I usually have no idea something happened until after it already changed. Open sourcing the core is a good first step toward that.
Flowly
@galdayan You've basically described why we built it. Capability was never the gap — it's that most agents act first and tell you later, if at all.
So Flowly does three things about exactly that: a live activity log you can watch it work in (not a recap after the fact), a full audit log to go back through, and per-action approvals so anything touching real data gets gated before it runs — not reviewed after it already changed.
Local models (Ollama / LM Studio / vLLM) cover the fully-offline part, and open-sourcing the core is the point — you shouldn't have to take our word that any of that's true, you can read it.
If you give it a shot, I'd genuinely want to know where it still falls short of "I knew before it happened."
Congrats on the launch! Flowly looks really interesting.
I wanted to try the mobile app, but it seems unavailable in Latin America. Is the mobile app currently limited to selected countries, or do you have plans to open availability for LatAm soon?
Flowly
@annki Thanks! And good catch — that's deliberate for now, not a bug. Some app-store regions require extra regulatory filings per country, and as a small team we launched with the regions we could get through first. LatAm is absolutely on the list — no date I can promise honestly yet, but it's a "when," not an "if."
Which country are you in? Genuinely helps us decide where to file next. In the meantime the desktop app and the open-source core work anywhere, no store involved — happy to help you get set up if you want to try it that way.
Runs natively on your own machine with your own model keys, plus a persistent memory of your world rather than just a chat log - that's a meaningfully different bet than most desktop agents. How does it decide what's worth remembering vs noise, is that tunable per-user?
Flowly
@dannyheng Great question — this is the part we've gone deepest on. Short version: memory here is a pipeline, not a transcript.
What gets in: during a conversation, durable facts get extracted (not the chat log itself), and each one lands as a governed record with its own confidence/trust score — dated, sourced, and tracked through a lifecycle (candidate → active → stale/superseded) instead of living forever as "true."
What kills the noise: a background pass reviews recent conversations after the fact, reconciles new candidates against what's already known — confident facts commit, uncertain ones go to a review queue for you. A separate consolidation pass merges duplicates and retires stale facts, so the memory self-corrects over time instead of silently rotting.
Tunable per-user — yes, three levers: (1) a commit mode — eager / selective / manual-review-everything, depending on how much you want to gate; (2) 👍/👎 on any memory, which actually retunes its trust score rather than just hiding it; (3) full inspect/edit/delete — memory panel in the apps, flowly memory list in the CLI. Nothing it knows is opaque to you.
It's the same bet as the rest of the product: the memory is yours, so you get the dials. Would love to hear how it holds up against your actual noise if you try it — that reconcile step is where real-world feedback matters most.
Flowly
Thanks! The core is genuinely quick to self-host — "curl -fsSL https://useflowlyapp.com/install.sh | bash", flowly setup to pick a model (local or hosted) and your channels, then flowly. Two minutes on a laptop, a Mac mini, or a $5 VPS, no account needed. flowly service install keeps it running in the background, so it's just… always there.
Making it yours happens two ways. One, you extend it: drop in your own Markdown skills, write Python plugins for tools/commands/channels, swap models or personas whenever. Two — and honestly this is the point — the memory learns your world as you use it, so over a couple of weeks it turns from "an agent" into your agent.
The power-user bits (some channels, sandbox tuning) take a little more, but the day-to-day is meant to feel boringly simple. Happy to help you get set up if you give it a shot.
The "memory of your world" being a model that tracks what changed and self-corrects is what would make me actually keep an agent around — a flat chat log always rots. Since data stays on-device but you've got Mac + iPhone in sync, how does that sync actually move: peer-to-peer / local network, or through a relay you host, and where does the memory live when one device is offline? And with the core open-sourced, is the memory store a documented local format I can inspect and back up, or an opaque embedded DB?
This is interesting a personal AI with memory could be really useful. Curious how people can manage or edit what it remembers?
Flowly