Skip to content

Releases

v0.6.0

Choose a tag to compare

@radys-s radys-s released this 18 Nov 16:37
a1272e0

Studio can now reach models through Refrakt, looks like a live dashboard, and is easier to use, especially on a phone and on Linux.

Highlights

  • Refrakt as a second gateway: models listed as refrakt/<provider>/<model> use the keys stored in Refrakt; set REFRAKT_GATEWAY_TOKEN to turn it on. Privacy rules follow the provider behind each model.
  • New look: a dark dashboard with motion that follows what a task is doing. All motion stops when the system asks for reduced motion.

Easier to use

  • Phones: the task list is visible again at 360 px, and "Podrobnosti" brings the details panel into view instead of opening it far below.
  • Linux: picking a project fills the repository from the project note on Linux too.
  • No gateway, no dead runs: when neither OmniRoute nor Refrakt answers, the new-task screen says what to start, and "Spustit" no longer creates a run that fails at once. The sidebar shows Refrakt's status next to OmniRoute's.
  • Clear decisions: when a run waits for you, the reason is explained in Czech and names the button to press. The original text is still under "Technický detail".
  • Correct Czech: counts read "1 záznam", "2 změněné soubory", "2 čekají na tebe" and so on.

Details: docs/ux-audit-2026-11.md and docs/progress.md.

v0.5.0

Choose a tag to compare

@radys-s radys-s released this 14 Nov 19:31
733479d

First tagged release of AI Bridge Studio: plan with Claude Code or Codex, implement with a cheap model, then verify and review the diff.

Highlights

  • Autopilot orchestration by default, with an optional second opinion from Claude or Codex. Work is routed only to models that don't train on your data.
  • A router that learns from outcomes: it moves models that often fail down the list and prefers models whose work gets accepted.
  • Overview: results, time to acceptance, fix rounds, decisions and usage for 7, 30 or 90 days.
  • Notifications and a pending count in the tab title. "Similar task" reuses earlier work, and the iteration checkpoint survives a crash.
  • Review: a more precise secret scanner, and the reviewer is never the same model through a different provider.
  • Providers: the same providers as Prizma, through OmniRoute. Settings link to each provider's API keys and have a Test button for every model.
  • Memory: continue a project from Obsidian, propose memory entries, and apply a change to a branch.
  • Simple UI: progress in 6 phases, with details on demand.
  • CI: Windows backend (ruff, strict mypy, pytest) plus frontend tests and build.