Added
- Special edition of Game mode Lorebook Keeper.
- Guides for all modes.
- QoL improvements to Lorebooks handling.
- Optional intuitive swipe navigation lets Conversation and Roleplay users move through rerolls with arrow keys or touch swipes, with an opt-in reroll-at-the-end shortcut.
- Roleplay chats can now optionally let characters create direct-message Conversation chats with hidden
[dm: ...]commands. - Lorebook entries can now be selected in bulk and copied or moved to another lorebook.
Fixed
- Various issues caused by the security tightening were fixed.
- Sidecar issues fixed.
- Improves the selfie regex, catching malformed commands.
- Fixed context trimming.
- MLX sidecar runtime installs the upstream
mlx-lmsource build so curated Gemma 4 MLX models can load on Apple Silicon. - Dry-run prompt preview now trims against manually configured preset Max Context Window values instead of only connection/model limits.
- Script custom tools now show their disabled state in the editor and fail safely when
CUSTOM_TOOL_SCRIPT_ENABLEDis off instead of silently disappearing from agent tool pickers. - Browser extensions now load under CSP through Blob module execution instead of eval, keeping extension support without adding
unsafe-eval. - Local sidecar runtime installation now works when the matching Admin Access secret is entered, even if
SIDECAR_RUNTIME_INSTALL_ENABLEDremains off. - Agent traffic now warns when the default agent connection may bill a provider. Agents explicitly set to Local Model are skipped with a visible warning when the sidecar is unavailable instead of silently falling back to a paid API connection.
- Chat attachments now wait for file reads, preserve files in manual group mode, and expose supported text files like JSON/Markdown/CSV to the model instead of silently dropping them.
- Fixed rolling in Game mode.
- Lorebook Keeper updates now receive existing entry content, and append structured new facts instead of replacing user-written lorebook text.
- Docker images now repair
/app/datavolume ownership before dropping to the non-root runtime user, preventingEACCESstartup failures during file-storage migration. - OpenAI-compatible local streams now accept stricter and looser SSE
data:formatting, Conversation mode visibly streams text again, and live reasoning chunks appear while a reply is still generating. - Expression agent sprite updates now repair stale character IDs from the current character name before dropping the expression, so existing characters keep their expressions mid-session.
- Stability AI image connections now test against Stability's account endpoint, fetch legacy v1 engines when needed, and generate through the correct v2beta Stable Image task endpoints instead of probing
/models. - Game mode party changes made from Chat Settings now sync to game metadata and carry into future sessions.
- NanoGPT GPT Image 2 requests now normalize image size to a supported pixel budget instead of forwarding too-small canvases.
- Conversation manual generations now share the autonomous in-progress guard, preventing async catch-up replies from duplicating the same user turn.