Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

readme: add instructions to build with MLX (#13733)

Jeffrey Morgan80f3f1bc25e1
diff --git a/README.md b/README.md
index 27bfd040d..bda6b4c34 100644
--- a/README.md
+++ b/README.md
@@ -48,7 +48,7 @@ ollama run gemma3
## Model library
-Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library 'ollama model library')
+Ollama supports a list of models available on [ollama.com/library](https://ollama.com/library "ollama model library")
Here are some example models that can be downloaded:
@@ -79,7 +79,7 @@ Here are some example models that can be downloaded:
| Code Llama | 7B | 3.8GB | `ollama run codellama` |
| Llama 2 Uncensored | 7B | 3.8GB | `ollama run llama2-uncensored` |
| LLaVA | 7B | 4.5GB | `ollama run llava` |
-| Granite-3.3 | 8B | 4.9GB | `ollama run granite3.3` |
+| Granite-3.3 | 8B | 4.9GB | `ollama run granite3.3` |
> [!NOTE]
> You should have at least 8 GB of RAM available to run the 7B models, 16 GB to run the 13B models, and 32 GB to run the 33B models.
@@ -260,6 +260,38 @@ Finally, in a separate shell, run a model:
./ollama run llama3.2
```
+## Building with MLX (experimental)
+
+First build the MLX libraries:
+
+```shell
+cmake --preset MLX
+cmake --build --preset MLX --parallel
+cmake --install build --component MLX
+```
+
+Next, build the `ollama-mlx` binary, which is a separate build of the Ollama runtime with MLX support enabled (needs to be in the same directory as `ollama`):
+
+```shell
+go build -tags mlx -o ollama-mlx .
+```
+
+Finally, start the server:
+
+```
+./ollama serve
+```
+
+### Building MLX with CUDA
+
+When building with CUDA, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with default architectures:
+
+```shell
+cmake --preset 'MLX CUDA 13'
+cmake --build --preset 'MLX CUDA 13' --parallel
+cmake --install build --component MLX
+```
+
## REST API
Ollama has a REST API for running and managing models.
@@ -422,7 +454,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
- [AppFlowy](https://github.com/AppFlowy-IO/AppFlowy) (AI collaborative workspace with Ollama, cross-platform and self-hostable)
- [Lumina](https://github.com/cushydigit/lumina.git) (A lightweight, minimal React.js frontend for interacting with Ollama servers)
- [Tiny Notepad](https://pypi.org/project/tiny-notepad) (A lightweight, notepad-like interface to chat with ollama available on PyPI)
-- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.)
+- [macLlama (macOS native)](https://github.com/hellotunamayo/macLlama) (A native macOS GUI application for interacting with Ollama models, featuring a chat interface.)
- [GPTranslate](https://github.com/philberndt/GPTranslate) (A fast and lightweight, AI powered desktop translation application written with Rust and Tauri. Features real-time translation with OpenAI/Azure/Ollama.)
- [ollama launcher](https://github.com/NGC13009/ollama-launcher) (A launcher for Ollama, aiming to provide users with convenient functions such as ollama server launching, management, or configuration.)
- [ai-hub](https://github.com/Aj-Seven/ai-hub) (AI Hub supports multiple models via API keys and Chat support via Ollama API.)
@@ -494,7 +526,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
### Database
- [pgai](https://github.com/timescale/pgai) - PostgreSQL as a vector database (Create and search embeddings from Ollama models using pgvector)
- - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
+ - [Get started guide](https://github.com/timescale/pgai/blob/main/docs/vectorizer-quick-start.md)
- [MindsDB](https://github.com/mindsdb/mindsdb/blob/staging/mindsdb/integrations/handlers/ollama_handler/README.md) (Connects Ollama models with nearly 200 data platforms and apps)
- [chromem-go](https://github.com/philippgille/chromem-go/blob/v0.5.0/embed_ollama.go) with [example](https://github.com/philippgille/chromem-go/tree/v0.5.0/examples/rag-wikipedia-ollama)
- [Kangaroo](https://github.com/dbkangaroo/kangaroo) (AI-powered SQL client and admin tool for popular databases)
@@ -637,6 +669,7 @@ See the [API documentation](./docs/api.md) for all endpoints.
- [llama.cpp](https://github.com/ggml-org/llama.cpp) project founded by Georgi Gerganov.
### Observability
+
- [Opik](https://www.comet.com/docs/opik/cookbook/ollama) is an open-source platform to debug, evaluate, and monitor your LLM applications, RAG systems, and agentic workflows with comprehensive tracing, automated evaluations, and production-ready dashboards. Opik supports native integration to Ollama.
- [Lunary](https://lunary.ai/docs/integrations/ollama) is the leading open-source LLM observability platform. It provides a variety of enterprise-grade features such as real-time analytics, prompt templates management, PII masking, and comprehensive agent tracing.
- [OpenLIT](https://github.com/openlit/openlit) is an OpenTelemetry-native tool for monitoring Ollama Applications & GPUs using traces and metrics.
@@ -645,4 +678,5 @@ See the [API documentation](./docs/api.md) for all endpoints.
- [MLflow Tracing](https://mlflow.org/docs/latest/llms/tracing/index.html#automatic-tracing) is an open source LLM observability tool with a convenient API to log and visualize traces, making it easy to debug and evaluate GenAI applications.
### Security
+
- [Ollama Fortress](https://github.com/ParisNeo/ollama_proxy_server)
diff --git a/x/README.md b/x/README.md
deleted file mode 100644
index 5af087380..000000000
--- a/x/README.md
+++ /dev/null
@@ -1,50 +0,0 @@
-# Experimental Features
-
-## MLX Backend
-
-We're working on a new experimental backend based on the [MLX project](https://github.com/ml-explore/mlx)
-
-Support is currently limited to MacOS and Linux with CUDA GPUs. We're looking to add support for Windows CUDA soon, and other GPU vendors.
-
-### Building ollama-mlx
-
-The `ollama-mlx` binary is a separate build of Ollama with MLX support enabled. This enables experimental features like image generation.
-
-#### macOS (Apple Silicon and Intel)
-
-```bash
-# Build MLX backend libraries
-cmake --preset MLX
-cmake --build --preset MLX --parallel
-cmake --install build --component MLX
-
-# Build ollama-mlx binary
-go build -tags mlx -o ollama-mlx .
-```
-
-#### Linux (CUDA)
-
-On Linux, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with the default Ollama NVIDIA GPU architectures enabled:
-
-```bash
-# Build MLX backend libraries with CUDA support
-cmake --preset 'MLX CUDA 13'
-cmake --build --preset 'MLX CUDA 13' --parallel
-cmake --install build --component MLX
-
-# Build ollama-mlx binary
-CGO_CFLAGS="-O3 -I$(pwd)/build/_deps/mlx-c-src" \
-CGO_LDFLAGS="-L$(pwd)/build/lib/ollama -lmlxc -lmlx" \
-go build -tags mlx -o ollama-mlx .
-```
-
-#### Using build scripts
-
-The build scripts automatically create the `ollama-mlx` binary:
-
-- **macOS**: `./scripts/build_darwin.sh` produces `dist/darwin/ollama-mlx`
-- **Linux**: `./scripts/build_linux.sh` produces `ollama-mlx` in the output archives
-
-## Image Generation
-
-Image generation is built into the `ollama-mlx` binary. Run `ollama-mlx serve` to start the server with image generation support enabled.