Changelog
Follow up on the latest improvements and updates.
RSS
improved
fixed
v0.15.3
Fixed
- Fixed an issue where overly long MCP server tool names could cause errors.
- Fixed security vulnerabilities.
Improved
- Updated and expanded the available model list.
improved
fixed
v0.15.2
What’s fixed
- Fixed an issue where DeepSeek tool calls could behave unexpectedly. #393
- Fixed a security issue related to Mermaid rendering.
- Fixed visual glitches in dark mode.
What’s improved
Improved MCP support with better tool execution and new support for resources and prompts.
Behind the scenes
Most of the recent work happened under the hood, it mainly focuses on stability and long-term reliability.
We rewrote the data layer, consolidating SQLite and LanceDB into PGlite, and reworked the MCP protocol layer to improve tool execution and enable resource- and prompt-level support.
We have ambitious plans for 5ire in 2026 — and recent work is about laying a solid foundation to make that possible.
fixed
v0.15.1
Hotfix: Fixed an error that caused knowledge-base file imports to fail.
improved
v0.15.0
Data layer fully rebuilt.
We’ve migrated the storage engine from SQLite + LanceDB to pglite.
This refactor doesn’t introduce new features yet, but it lays a solid foundation for upcoming improvements and enhanced capabilities. Updating is recommended to stay compatible with future releases.
Update Strategy Change
Auto-update disabled.
Future versions will no longer install automatically.
Instead, you’ll receive a notification when a new version is available and can update manually at your convenience.
📌 Notes
Existing data will be migrated automatically during the upgrade.
improved
fixed
v0.14.1
Grok & Moonshot Upgraded, Bugs Squashed
- Updated built-in models (Grok, Moonshot)
- Fixed Azure provider API key header issue
- Fixed Gemini tool invocation bug
new
improved
fixed
v0.14.0
What’s New
- You can now set approval rules for tools
- MCP Server now supports prompt templates
Improvements
- MCP Prompts and Tool outputs are automatically converted into model-friendly formats.
- Better compatibility when working with non-text content across providers like Ollama, Google, Anthropic, and Mistral.
Fixes
- Fixed mismatches between model form inputs and their configuration.
- Resolved an issue where MCP Tools and Prompts couldn’t be used if the MCP Key included a hyphen (“-”).
- Security Patch: Fixed an XSS vulnerability caused by ECharts rendering.
⚠️ Since this update includes a critical security fix, we strongly recommend upgrading to the latest version as soon as possible.
new
improved
fixed
v0.13.2
New
- Added built-in provider: Perplexity (citation links not yet shown in UI)
- Added Grok-4 as a built-in model
- Added optional API Key setting for Ollama
Improved
- Improved UI on macOS for a more native experience
- Enhanced KaTeX rendering for array formulas
- Improved performance when handling large text outputs
Fixed
- Fixed a white screen issue when creating a new chat right after deletion
- Fixed inability to clear Proxy settings in provider configuration
- Fixed missing name validation when saving models
new
improved
fixed
v0.13.0
New Features
- Added support for setting proxies for providers
- Added built-in provider: 302AI
- Added right-click menu to folder and chat navigation in the sidebar
Improvements
- Window styling adapted for both Windows and Linux systems
Bug Fixes
- Fixed issue where switching conversations mid-reply could cause errors
- Attempted fix for MCP Server not accepting self-signed certificates (requires setting `NODE_EXTRA_CA_CERTS on app launch)
fixed
v0.12.1
hotfix
Fix the issue with the incorrect temperature value during initial loading.
improved
fixed
v0.12.0
🚀 Optimizations
- Added more built-in models for 5ire Provider
- Aligned 5ire model capabilities with official standards, supporting vision and tools
- Enhanced error handling mechanisms for tool invocation
- Upgraded Baidu services to new API version (using API Key authentication)
🐞 Bug Fixes
- Fixed failure in switching new chat models
- Resolved application crash caused by bulk knowledge imports
Load More
→