Repository navigation
What's Changed
Includes the improvements from 4.19.0, whose release publication did not complete.
Added
- Provider-owned image models and localized built-in tool descriptions.
- Multilingual memory retrieval, archiving, and a memory management panel.
Changed
- Refined search provider settings, provider selection, and API-key actions.
- Bundled memory is disabled by default, and its settings stay hidden until enabled.
- Updated Pi model and agent libraries to 0.87.1, verified against the official
npm registry on 2026-09-29. Retained the local Responses compatibility and
native-tool patch because upstream still lacks the required behavior,
including tool-call ID replay for non-OpenAI providers.
Fixed
- Preserve literal edit-tool replacements, imported tool-result errors, blank
context lines in patches, and insertion-only patch hunk ordering. - Wait for plugin ZIP writes to settle before cleaning up failed installations.
- Keep dynamic prompt variables fresh, record resolved run models, and preserve
memory records during upgrades and concurrent writes. - Confine local filesystem operations, reject nested symlinks during recursive
copies, and cap tool output without splitting UTF-8 characters. - Synchronize delayed fullscreen state.
Install
Two editions. Install either, or both — they share the same ~/.llm-space data, so switching keeps your threads and settings.
- LLM Space — uses the system WebView. Small download (~27 MB), light on memory and battery.
- LLM Space Performance — embeds its own rendering engine (~130 MB). Rendering stays consistent across macOS versions, and usually performs better.
Download a DMG below. Existing installs update in place.
Full changelog: https://github.com/deer-flow/llm-space/blob/v4.19.1/CHANGELOG.md