LoginSign Up
← Back to all sparks
C

Cherry Studio

AI-ASSISTANTS
Velocity5.0

Desktop AI client supporting multiple LLM providers and knowledge bases.

www.cherry-ai.com ↗

The v2 rewrite has shipped; Cherry Studio is back to patch releases.

desktop-ai-clientv2-rewritedata-migrationllm-providersrelease-train
◆Current state
Cherry Studio spent late July running a v2.0.0 release train - three betas and five release candidates inside two weeks - to land a rewrite that had merged into main while v1 code still sat alongside it. The August entry is v2.0.6, a single Files-page bug fix, which puts the product past the rewrite and into ordinary patch cadence. The feed never carried a v2.0.0 GA note: it jumps from rc.5 on 4 August straight to v2.0.6 on 17 August.
◆Where it's heading
The release train's substance was migration safety rather than new capability - preserving model endpoint routing, stopping table-recreate migrations from silently deleting child rows, keeping Claude session and workspace continuity, restoring guarded v1 style migration. Provider work continued underneath it, with Gemma 4 thinking in Ollama, a Radeon Cloud integration, and a configurable default endpoint. The priority through the whole train was getting existing users across the v1/v2 boundary with their data and settings intact.
◆Prediction
With v2 out and the patch stream started, the next entries should shift back from migration repair to provider and agent features - the strand that kept moving quietly through the rc series.

◆Recent moves

  1. 2d ago

    Files page keeps the upload button in every category

    A single fix: the Files page hid the upload button outside the All files view, so uploading from an Image or Document category meant navigating back first. It is the first patch after the v2 train and reads like the start of routine maintenance rather than a continuation of the rewrite.

    View source ↗
  2. 14d ago

    rc.5: Gemma 4 thinking in Ollama, anchor rail navigation

    The last candidate before v2 carried real additions among the fixes - Gemma 4 thinking mode through Ollama, a minimal tick-rail for message anchor navigation, save-as actions in image preview - alongside streamed code-block and file-processing repairs. It is the point where the train stopped being purely defensive.

    View source ↗
  3. 17d ago

    rc.4: Radeon Cloud provider and configurable endpoints

    Adds a Radeon Cloud provider integration and lets users change a provider's default endpoint, plus unified file and artifact preview routing and a converged semantic color contract. Provider breadth is the one axis Cherry Studio kept extending while the rest of the train handled migration.

    View source ↗
  4. 19d ago

    rc.3: knowledge, export, and accessibility fixes

    Accessibility labels on page-zoom controls, an Obsidian export dialog that stays open on failure, readable names for saved knowledge messages, and a shared resource-group creation dialog. Cleanup work with nothing a user would call a new capability.

    View source ↗
  5. 20d ago

    rc.2: migration fixes protect routing and child rows

    The most consequential candidate of the train: model endpoint routing preserved across migration, Claude session and workspace continuity kept for agents, and table-recreate migrations stopped from silently deleting child rows. These are the failures that would have made the v1-to-v2 jump lossy, caught before GA.

    View source ↗
  6. 21d ago

    rc.1: packaging and onboarding fixes

    Windows portable builds get sqlite-vec, x64 macOS builds ship sharp, onboarding filters non-chat models, and migrated mini-apps keep their icons. Build and first-run plumbing for the v2 cutover.

    View source ↗