v9.1.1 is the release where everything clicked. We listened to feedback from real users actually running Skales on consumer hardware, using it with local models, and managing workflows. The result is a system that's stable, smart, and fast.

Local Model Fixes: The Big One

Ollama and local models were causing timeouts and hanging chats. Here's what we fixed:

Ollama/local model tools disabled by default — local models are slow at tool use. We disabled them entirely to eliminate random timeouts on consumer hardware.
Tool slider (0-70 range, step 5, default 15) — for users who want to enable tools with local models, you now control exactly how many tools get passed to the model. Default is 15 for stability.
Fast-fail retry (10s timeout) — if a local model times out on a tool call, we immediately retry without tools. Gets you a response instead of hanging forever.
Chat hard-kill extended to 200s — local providers get more time to think before we force-quit a stuck chat.

The philosophy: local models are fast at reasoning, slow at tool use. We optimize for that reality.

Advisor Strategy Fixed

This was broken for weeks and nobody told me. The advisor strategy uses Opus (or GPT-4) for planning and Sonnet (or GPT-3.5) for execution. Sounds great. But the advisor logic only existed in processMessageWithTools(), which chat mode never called.

Fixed. Now the advisor actually runs, plans with a heavy model, executes with a light one, and thinks strategically before acting.

Agent Skills: Save, Import, Mention

You can now save agent skills directly from a chat. Run a workflow, refine it, save it. Skills persist to disk. We also built a bulk import from GitHub — drop a GitHub raw URL and import 50 skills at once.

In chat, use @skillname to load a specific skill into context. Perfect for complex multi-step tasks.

Studio: API Key Sharing & Cloud Video

Studio now shares API keys from your main settings. No need to re-enter them. We also added cloud video generation support:

Google Veo (text-to-video, 1280x720 native)
Kling AI (text-to-video)
Runway (Gen-2, Gen-3 alpha)
MiniMax (video generation)
Seedance (video generation)

Use any of these in Studio with your own API key. Generate videos from prompts.

WordPress: Major Overhaul

The WordPress integration got huge:

Test Connection crash fixed — no more crashes when testing your connection.
Full-width CSS v2 — better rendering of WordPress content in the client.
AI Command Bar rewrite — faster, more responsive.
Elementor exits beta — fully supported with 5 templates included.
Plugin v1.1.0 — stable, well-tested WordPress plugin ready for production.

13 New Slash Commands (24 Total)

Slash commands are shortcuts for quick actions. We added:

/memory — view and manage stored memories
/skills — list and load agent skills
/provider — switch AI provider in chat
/version — check your Skales version
/export — export conversation to markdown
/theme — toggle light/dark mode
Plus 7 more for common workflows

Identity Maintenance Silent Mode

Identity Maintenance kept asking for approval every time it ran. We fixed that. It runs silently now, consolidates your memory in the background, and doesn't interrupt you.

Tilde Expansion for File Tools

File paths with ~ (home directory) now expand correctly everywhere. ~/Documents/project.md works in all file tools.

Calendar Sync Status Indicators

You can now see at a glance whether your calendar is synced and healthy. Red/yellow/green status in settings. If something's wrong, you see it immediately.

The Result

v9.1.1 feels like the first stable release. Local models work. Skills are real and usable. WordPress integration is production-ready. Slash commands save you clicks. It's the completion of the v9.0 foundation work.

Try it out. Use it daily. It'll just work.

Download Skales — free on Windows, macOS, and Linux.

— Mario 🦎