v11.3.0 - Together 🤝

Together is about one thing: Skales is no longer alone. Pair with another person's Skales and work a shared plan, hand tasks to your own other devices, and watch every answer stream in live while it's being written.

🤝 Teams (Beta): two people, two agents, one plan

Pair your Skales with another PERSON's Skales, end to end encrypted, confirmed by name on both sides. You chat in a shared thread, and the You/Agent switch lets your agent answer with this machine's full tool set.
The team chat carries a shared plan. Both of you add checklist items and assign each one to a machine, or hit "I'll take everything" to hand the whole list to one device.
Nothing runs until BOTH of you approve. Any edit to the list, including a takeover, voids both approvals, so nobody can swap tasks after sign-off. Once armed, each computer works through exactly its own items with its own tools and keys, attaches results, and posts each outcome into the thread.
Built local-sovereign: a teammate's words are never instructions, a received plan can never execute anything by itself, and your approval can never be forged remotely.

🐝 Swarm grows up to Beta: your other devices, one workforce

/swarm in any chat picks the best free device, /swarm @name targets one, and a mode prefix sets how it runs there: code: (coding agent), plan: (read-only), auto: (fully autonomous).
Results come back INTO the chat you sent from, plus Notifications, and the device doing the work shows a 🐝 badge with the sender's name.
Manually added devices (fixed IP, Tailscale) survive restarts and no longer fall to offline after 60 seconds; a health loop keeps every peer's status fresh.
Receiving stays opt-in per device and both sides need the same swarm secret. A missing secret is flagged on the page instead of failing silently.

⚡ Answers stream in live

Replies appear token by token instead of all at once, on OpenAI-compatible cloud providers (OpenAI, OpenRouter, Groq, DeepSeek, custom) and on local models (Ollama, LM Studio, llama.cpp). The biggest felt-speed difference, especially locally. Anthropic and Gemini native connections still deliver in one piece for now.
Thinking models show their reasoning live, collapsed and stable while it grows, no flicker, no width jumps.
Streaming stays in one bubble, scrolling up while Skales types finally sticks, and the jump-down arrow reliably appears above the chat and below every popup.

🌍 Discover comes alive

The "Your AI wants to share" cards actually arrive now, and finishing real work nudges a fresh suggestion instead of waiting hours. You approve or skip, nothing posts on its own.
Your weekly Wrapped lands in the feed as a picture, identical pixels on every screen, and any image in your gallery can be shared with the same compress-and-review pipeline.

🧭 Skales knows itself now

Ask "where do I set the voice provider" and get the right answer.
What's New lives inside the app: scrollable release notes under Settings -> Advanced -> Updates AND on the Update page itself, fetched live so even older installs read about the newest version.

📎 Files behave

Attaching several files no longer dumps the second one as raw text into your message. Every file shows as its own compact chip, your question below.
Drop a zip into the chat. Archives (zip, tar, gz, 7z, rar up to 25 MB) are saved to your Workspace and attached as a reference, then just ask Skales to extract them or read specific files from them.

⏰ Schedules: the follow-up

Schedules that worked before v11.2.7 fire again. The executor and the queue disagreed about which toggle authorizes them; now either Always-On or Autonomous Mode runs your schedules, on every machine.
A schedule paused after repeated failures shows as paused with the reason in its log, instead of displaying Active while being skipped forever. One toggle re-arms it, and stale failure counters from the old scheduler era are cleared once automatically.

💚 Friend Mode cannot silently die again

Undelivered check-ins are no longer recorded as sent, failed deliveries retry after about 15 minutes, and the quiet phase means one message per day instead of silence forever.

🧠 Better with every model

Per-model profiles ship ON by default, with new built-ins for Phi and GPT-OSS and refreshed tool hints everywhere. Community library: github.com/gonemedia/llm-profiles.
Thinking traces from Kimi-style models stop bleeding into the answer, the thoughts land in the collapsed trace where they belong.

✨ And the rest

• The update restart is silent now: no error flash for a split second before the app closes.
• Team and Swarm pages introduce themselves on first visit, one "Got it" and the explainer never returns.
• The "Skales can proactively send GIFs" switch does something now, sparingly and never on serious topics. (fixed)
• The get-to-know-you question can appear after a normal chat answer, same protections as before.
• Your dream diary is visible on the Memory page, and what Skales learns about you now genuinely shapes its replies.
• No more decorative emoji row above replies.
• Everything new is translated in all 12 languages.

Update: Skales updates itself, or grab the installer below.