A local-first AI harness for Windows, macOS, Linux, Android & iOS
Double-click. Run. Done. No Docker. No terminal. No signup.
Desktop 12.8.4 · Mobile 2.8.4
Skales is a harness, not a wrapper. A wrapper puts a chat window in front of someone else's API. A harness is the part that makes a model actually do things: the agent loop, the tool layer, the context and memory handling, the provider abstraction, and the per-model tuning that keeps tool calls landing across models that behave nothing alike.
Concretely, that is 257 tools on one turn budget, 26 providers behind one interface, a bundled local inference server, and a profile layer that adjusts sampling and tool naming per model family. The chat window is the surface. The harness is the product.
Local-first by default. Your data stays in ~/.skales-data on your machine.
Bring your own keys, or run with no key at all. No Docker. No terminal. No signup.
Built from 6 to 60+ — usable by a kid, approved by a grandmother. Mainstream surface, power-user depth underneath.
"From every tool I've tested in this space, I haven't found one that delivers intelligence without complexity, a companion instead of a tool, or value without hype. Skales has the foundation to tell that story." — @v33-kind, Community Contributor
| Repo | What | License |
|---|---|---|
| skales | Desktop harness (Windows, macOS, Linux) | BSL 1.1 |
| skales mobile | Android + iOS — both live | BSL 1.1 |
| devkit | CLI, developer docs, examples | MIT |
| wordpress | WordPress Connector Plugin | MIT |
| llm-profiles | Per-model tuning profiles for reliable tool calling | MIT |
| api-connectors | Keyless REST connector templates | MIT |
Skales ships its own llama.cpp server and starts it itself. One process, a directory of models, hot-swapping between them. No Ollama install, no Docker, no terminal — the download button is the whole setup.
The catalogue holds 55 downloadable assets: 18 local LLMs, their vision projectors, and the rest of the stack that makes a machine self-sufficient —
- 🖼️ Image generation on device via stable-diffusion.cpp — DreamShaper 8 LCM, Stable Diffusion 1.5, Z-Image Turbo, FLUX.2 klein
- 🎙️ Speech to text on device via Whisper
- 🔊 Text to speech on device — Piper voices run through sherpa-onnx, in eight languages
Everything above works with no API key and no network call.
Prefer a hosted model? Skales IQ is a free trial tier that answers instantly with fast open models and no key at all, so a fresh install is useful before you have signed up for anything. It is capped; when it runs out, Skales tells you and points at the place to add your own key.
🧰 257 tools — files, shell, git, browser, email, calendar, messaging, documents, images, memory, and the rest, loaded on the same turn
🌍 26 providers — OpenAI, Anthropic, Google, Groq, DeepSeek, xAI, Mistral, Together, MiniMax, Moonshot, GLM, Qwen, Hunyuan, GigaChat, Nvidia NIM, Cloudflare, Hugging Face, OpenRouter, Atlas Cloud, and the local ones: Ollama, LM Studio, Unsloth Desktop, Skales Local, plus any OpenAI-compatible endpoint
💻 Code — an agentic coding surface with a 3-panel diff GUI, session history and undo
🏢 Organization — build an AI company with departments, specialized agents and team leaders
🤖 Agents — custom agents with opt-in per-agent memory: each finished task leaves a lesson the next run reads back
🎛️ Cockpit — goals, tasks and schedule in one screen, with a background heartbeat that works the queue every few minutes
🦁 Lio AI — block-based code builder for newcomers, generating HTML/CSS/JS/Python into a sandboxed preview
🎨 Studio & Flow — design, image, video, voice and music generation, with Brand Kit and PowerPoint/Word/PDF export
📝 WordPress lives in the chat — 47 tools behind the connector, so posts, pages, media and settings are just things you ask for
🌐 Built-in browser — the agent navigates, clicks and extracts
🖥️ Computer use — screenshot, click, type, scroll
📅 Planner + calendar sync — Google, Apple, Outlook
🧠 Memory + dreaming — persistent memory with 3-phase overnight consolidation
🗣️ 12 languages — EN, DE, ES, FR, RU, PT, KO, ZH, JA, VI, HR, TR
The voice inside Skales has a name and a face. Iris opens in her own frameless window — wake word, listening, speaking — and carries on the same session as the chat you handed to her, while the rest of Skales stays where it was. She ships today, marked Beta.
Iris Orbit becomes its own product later: irisorbit.com
BSL 1.1 source-available today, Apache 2.0 on 2030-04-19.
Personal, educational, and internal business use = free forever.
SaaS, hosting, bundling, or resale = commercial license.
Made by Mario Simic in Vienna 🇦🇹
