Skales

Offline local-model app · Last verified July 22, 2026

Skales vs. Jan

The purest offline model runner vs. a local-first agent that also does the work.

Jan (roughly 43,700 GitHub stars as of July 22, 2026, Apache-2.0, by Menlo Research in Singapore) is the cleanest answer to one specific requirement: run models on my own disk, with no account, no network and no compromise. It downloads GGUF models from HuggingFace, runs them on its own llama.cpp engine, and exposes an OpenAI-compatible server on localhost so other apps can use it as a backend. Apache-2.0 makes it the most permissively licensed product on this page.

Skales shares the conviction and adds the other half. It also runs fully offline through Ollama or LM Studio - and on the phone, through a model on the device - but its point is what happens after the answer: 183 tools acting on your files, browser, calendar and mail, goals that keep running in the background, a coding agent, Flow and a media studio. If Jan is the engine, Skales is the vehicle. They are not enemies; you can point Skales at Jan's local server.

Skales Jan What this means for you
Purpose Local-first agent that does work on your machine Local model runner and chat app Different jobs. Both start from the same privacy conviction.
Offline Yes - Ollama, LM Studio, any local endpoint, and an on-device model on the phone Yes - its own llama.cpp engine, downloads and runs GGUF directly Genuine credit: model management is Jan's home turf.
Acting on your computer 183 tools: files, browser, computer use, calendar, email, messengers MCP integration for tool use Built in vs. brought in.
Background work Goals with the chat closed, recurring schedules, OODA self-correction Interactive sessions Skales keeps going when you stop.
Mobile Native app on both stores, standalone, runs a model on the device None - the app-store epic was closed as not planned A phone agent exists on one side only.
Creation Flow + Studio: image, video, voice, music, design Not in scope - text and chat Jan stays deliberately narrow.
Interoperability Speaks to any OpenAI-compatible endpoint - including Jan's Serves an OpenAI-compatible endpoint on localhost:1337 These two compose. Run Jan as the engine, Skales as the agent.
License BSL 1.1 → Apache 2.0 in 2030 Apache-2.0, no restrictions Genuine credit: theirs is the freer licence today.

Choose Skales if…

  • You want the agent to do things, not only to answer - across your files, mail and calendar.
  • You want background goals, a coding agent, Flow and media generation in one app.
  • You want the same agent on your phone, including offline.

Choose Jan if…

  • Your requirement is model management: browse, download, run and serve GGUF models locally.
  • You want a permissive Apache-2.0 licence with no commercial conditions at all.
  • You want the smallest possible surface: local inference and a chat window, nothing else.

Common questions

Is Skales better than Jan?

They solve different problems. The purest offline model runner vs. a local-first agent that also does the work. Skales is the right choice if: You want the agent to do things, not only to answer - across your files, mail and calendar. Jan is the right choice if: Your requirement is model management: browse, download, run and serve GGUF models locally.

Is Skales free while Jan is also free?

Skales is free for personal, educational and internal business use (BSL 1.1, converting to Apache 2.0 in 2030). You bring your own AI key or run local models for a total cost of zero.

Can I migrate from Jan to Skales?

Skales imports settings and conversations from ChatGPT, Claude, Copilot, Gemini, OpenClaw, Hermes and Cherry Studio via Settings → Import from Another Tool.

Sources

Spotted something outdated? Mail request@skales.app - we fix comparison pages within days.