Cline shipped Desktop v0.0.49, a large update adding a Changes pane with per-turn diffs, a project files panel, an integrated terminal, keybindings, and a Local Models settings page for Ollama and LM Studio. The release also fixes session attach/duplicate-run issues, large-transcript resume failures, and a git process explosion, and refreshes the model catalog with new providers and defaults.
Cline released CLI v3.0.74 and SDK v0.2.0 on GitHub, twin reliability updates to its coding agent. Sessions that reconnect to in-progress runs now reattach instead of spawning unstoppable duplicates, model-request retries dropped from 5 to 3, and turns dropped after reasoning-only streaming are retried. Both releases also refresh the model catalog, adding Liquid Inference and Munito AI providers and defaulting most providers to Kimi K3.1 Pro, MiniMax M3.1 Flash, or DeepSeek V4.1 Flash 1017.
Cline released CLI v3.0.74 and SDK v0.2.0 on GitHub, twin reliability updates to its coding agent. Sessions that reconnect to in-progress runs now reattach instead of spawning unstoppable duplicates, model-request retries dropped from 5 to 3, and turns dropped after reasoning-only streaming are retried. Both releases also refresh the model catalog, adding Liquid Inference and Munito AI providers and defaulting most providers to Kimi K3.1 Pro, MiniMax M3.1 Flash, or DeepSeek V4.1 Flash 1017.
MiniMax launched M3.1 Flash on October 19, 2026, confirming that 'Space Bunny' — an anonymous model that topped OpenRouter and OpenCode leaderboards for 10–12 days with 200T+ tokens served — was an earlier build of the same model. The final model ships free on OpenRouter and OpenCode for 72 hours, is available day-zero in Cline and Ollama, and targets long agentic coding runs where cost accumulates.
MiniMax launched M3.1 Flash on October 19, 2026, confirming that 'Space Bunny' — an anonymous model that topped OpenRouter and OpenCode leaderboards for 10–12 days with 200T+ tokens served — was an earlier build of the same model. The final model ships free on OpenRouter and OpenCode for 72 hours, is available day-zero in Cline and Ollama, and targets long agentic coding runs where cost accumulates.
Cline shipped Desktop v0.0.44, adding interactive inline Mermaid diagram rendering in chat and automatic Cline Hub session reconnection. The release also fixes broken local/self-hosted providers, Windows MCP server loading, and custom Anthropic base URLs.
Cline shipped Desktop v0.0.44, adding interactive inline Mermaid diagram rendering in chat and automatic Cline Hub session reconnection. The release also fixes broken local/self-hosted providers, Windows MCP server loading, and custom Anthropic base URLs.
A developer details their migration from Ollama to LM Studio for running local LLMs on Windows, citing faster inference via llama.cpp, VRAM-fit badges, GUI parameter control, MCP integrations, and the ability to reuse existing Ollama models. The post reflects growing community interest in LM Studio's features for local AI workflows.
A developer details their migration from Ollama to LM Studio for running local LLMs on Windows, citing faster inference via llama.cpp, VRAM-fit badges, GUI parameter control, MCP integrations, and the ability to reuse existing Ollama models. The post reflects growing community interest in LM Studio's features for local AI workflows.
Sebastien Burel released KaozKit, an MIT-licensed Swift library that runs JavaScript LLM agents inside Mac apps fully on-device. Agents execute in XS, a microcontroller-oriented JS engine, sandboxed so they see only host-provided functions. It supports MLX, Apple Intelligence's on-device model, Ollama, and LM Studio.
Sebastien Burel released KaozKit, an MIT-licensed Swift library that runs JavaScript LLM agents inside Mac apps fully on-device. Agents execute in XS, a microcontroller-oriented JS engine, sandboxed so they see only host-provided functions. It supports MLX, Apple Intelligence's on-device model, Ollama, and LM Studio.