Releases·Tool Release
ollama/ollama v0.34.3: thinking controls in /api/show, MLX vision support
Ollama released v0.34.3, making GET /api/show advertise each model's thinking levels and default, using glm-5.3-flash:cloud as the example. The release also adds MLX-based Apple Silicon support for Nemotron H vision models and fixes macOS window reopening.
2 engagement1 sourceSat, Sep 19, 2026, 12:02 AM
AI Overview
- `GET /api/show` now advertises each model's thinking levels and default, e.g. `glm-5.3-flash:cloud` defaults to `max`
- Nemotron H vision models gain Apple Silicon support via MLX
- The macOS app no longer reopens closed windows on activation
Entities
Related
Releases·Tool Release
ollama releases v0.34.1 with MLX memory and cache fixes
Ollama published v0.34.1, a patch release updating its MLX runner and llama.cpp backends. Changes focus on memory handling: the mlxrunner now evicts prefix cache snapshots and waits for evicted runners and free system memory before loading the next model. The release also raises the token repeat limit to 100 and fixes a ChatGPT model selector spacing bug in the app.
Ollama published v0.34.1, a patch release updating its MLX runner and llama.cpp backends. Changes focus on memory handling: the mlxrunner now evicts prefix cache snapshots and waits for evicted runners and free system memory before loading the next model. The release also raises the token repeat limit to 100 and fixes a ChatGPT model selector spacing bug in the app.
14 engagement1 sourcegithubMon, Sep 14, 2026, 10:14 PM
Releases·Tool Release
Ollama ships v0.33.x releases with MLX runner updates and Qwen3.8 Flash Next support
Ollama released three versions in a week: v0.33.0 (Aug 21), v0.33.1 (Aug 26), and v0.33.2 (Aug 27). The updates add a Claude desktop app, MLX runner improvements including structured output support and Qwen3.8 Flash Next support, plus app and proxy fixes. Practitioners running local models on Mac benefit from Metal GPU timeout fixes and faster structured output.
Ollama released three versions in a week: v0.33.0 (Aug 21), v0.33.1 (Aug 26), and v0.33.2 (Aug 27). The updates add a Claude desktop app, MLX runner improvements including structured output support and Qwen3.8 Flash Next support, plus app and proxy fixes. Practitioners running local models on Mac benefit from Metal GPU timeout fixes and faster structured output.
Updated Wed, Sep 2, 2026, 12:11 AM — v0.33.2 released Aug 27 with dark mode restore, proxy catalog-change handling, and macOS handoff sync.
04036 engagement4 sourcesgithubFri, Aug 21, 2026, 10:52 PM
Products·building
HR Policy RAG Assistant built locally with Ollama, Qdrant, FastAPI, and Streamlit
An end-to-end locally hosted Retrieval-Augmented Generation system that answers questions based on HR policy documents. It uses Ollama for local LLM inference and Qdrant as the vector store, so sensitive employee questions never leave the organization's infrastructure. Aimed at HR teams or employees who need quick, accurate answers about internal policies.
An end-to-end locally hosted Retrieval-Augmented Generation system that answers questions based on HR policy documents. It uses Ollama for local LLM inference and Qdrant as the vector store, so sensitive employee questions never leave the organization's infrastructure. Aimed at HR teams or employees who need quick, accurate answers about internal policies.
0 engagement1 sourcerssFri, Sep 4, 2026, 07:09 AM
Releases·Tool Release
Ollama v0.32.6 adds MLX speculative decoding, OpenAI-compatible streaming
Ollama released v0.32.6, improving Apple GPU performance via automatic speculative decoding with Qwen3.5's MTP head and aligning its streaming API with OpenAI's wire format. The update also fixes finish_reason reporting and adds cloud model fallback for kimi-k3.
Ollama released v0.32.6, improving Apple GPU performance via automatic speculative decoding with Qwen3.5's MTP head and aligning its streaming API with OpenAI's wire format. The update also fixes finish_reason reporting and adds cloud model fallback for kimi-k3.
11 engagement1 sourcegithubTue, Aug 4, 2026, 06:49 PM
Releases·Tool Release
block/goose v1.42.0 adds Ollama cloud provider and per-message usage stats
block/goose v1.42.0 introduces an OllamaCloudProvider with dynamic model discovery and context limits, along with per-message usage statistics UI showing tokens, cost, TTFT, and tok/s. The release also includes session-level cost tracking and improved project navigation.
block/goose v1.42.0 introduces an OllamaCloudProvider with dynamic model discovery and context limits, along with per-message usage statistics UI showing tokens, cost, TTFT, and tok/s. The release also includes session-level cost tracking and improved project navigation.
0 engagement1 sourcegithubMon, Jul 13, 2026, 07:33 PM