Run Open WebUI, Ollama, and Qwen3 8B on a MacBook Pro M4 16GB
A reproducible two-layer local chat stack for a 16GB M4 Mac: Ollama loads Qwen3 8B Q4 with Metal, while Open WebUI provides persistent conversations and a browser interface. Compatibility is documented, but this exact complete stack still needs its first submitted run report.
Evidence status
Ollama, Open WebUI, and the Qwen3 8B artifact each have documented compatibility with this path. LocalAIRun has not yet received a completed report for the full Open WebUI + Ollama stack on the exact 16GB M4 Mac, so expected values below are clearly labeled planning signals rather than first-hand results.
Requirements
- A working Ollama Qwen3 8B baseline
- Docker Desktop or a supported Python 3.11/3.12 environment
- Persistent local storage for Open WebUI
- Port 3000 or 8080 available
Stack
- Web application
- Open WebUI · Pinned stable release recommended
- Runtime
- Ollama · Current stable release
- Model artifact
- Qwen3 8B Q4_K_M · GGUF Q4_K_M
Use a named persistent volume and pin a release for a durable installation.
Ollama remains responsible for loading Qwen3 and using Metal.
Verify the Ollama-only recipe before adding the web layer.
Before you run it
Expected result and evidence boundary
Qwen3 8B Q4_K_M artifact metadata
LocalAIRun artifact estimate with OS and application headroom
Recipe baseline; larger contexts use more unified memory
M4 16GB hardware profile; not yet verified by this complete recipe
A planning signal is not a test result. The report count changes only after a completed run is submitted and accepted for this exact recipe.
Manual mode
Step-by-step execution
Use the same safeguards as the Skill
Check the machine first, use only the linked official sources, confirm large downloads and file locations, and do not bypass macOS security. Mark a step complete only after its check passes.
1. Verify Ollama first
- •Complete the Ollama Qwen3 8B baseline before installing another layer.
- •Confirm the Ollama API responds on localhost.
Check: Qwen3 works directly through Ollama.2. Choose and install Open WebUI
Official instructions →- •Use the official Docker path for the most reproducible installation, or the documented Python 3.11/3.12 path.
- •Create persistent storage and do not use the dev image for real data.
Check: Open WebUI opens locally and retains its configuration after restart.3. Connect the local runtime
- •Configure Open WebUI to reach the existing Ollama service.
- •Select Qwen3 8B and keep the baseline context unchanged.
Check: The Qwen3 model appears and responds through Open WebUI.4. Check persistence and report
- •Restart Open WebUI and confirm the local account and conversation persist.
- •Record image or package version, installation method, model response, and any connectivity error.
Check: The web UI survives restart and completes the fixed prompt.
Common failure points
Troubleshooting this exact stack
A container is trying to use its own localhost instead of the Mac host
Use the official Open WebUI Docker connection guidance and verify the Ollama API from the host before changing model settings.
The expected Qwen3 tag is not pulled or Open WebUI is connected to a different Ollama endpoint
Confirm the exact Ollama model tag directly, then refresh the Open WebUI model connection.
Context, parallel requests, or another application is consuming unified memory
Return to the 8K baseline, close memory-heavy applications, and test one request at a time.
Open WebUI storage was not mounted persistently
Stop before adding real documents, create the documented persistent volume or data directory, and repeat the restart check.
Baseline settings
- context length
- 8192
- open webui storage
- persistent
- open webui release
- pinned
Comparable test
Prompt
Summarize the difference between a model runtime and a chat interface in five bullet points.
Seed: 20260816
Did it run on your hardware?
Success and failure reports both improve this recipe's evidence level.
Test the layers separately
Start with the runtime-only baseline before adding another application layer.