ollama-coding-profiles

Cross-platform Ollama launch profiles and coding-focused Modelfiles for macOS and Linux, tuned especially for Apple Silicon laptops and local agent workflows like OpenCode and Aider.

This repo is built for people who want to:

• start ollama serve with repeatable profiles instead of retyping env vars
• keep separate presets for stable, daily coding, OpenCode, Aider, and max-context use-cases
• version control recommended Modelfiles for programming workloads
• use the same wrappers on both macOS and Linux

What is inside

.
├── README.md
├── .gitignore
├── bin/
│   ├── ollama-profile
│   ├── ollama-profile-m1-stable.sh
│   ├── ollama-profile-m1-balanced.sh
│   ├── ollama-profile-m1-opencode.sh
│   ├── ollama-profile-m1-aider.sh
│   ├── ollama-profile-m1-max-context.sh
│   ├── ollama-profile-private.sh
│   └── ollama-profile-debug.sh
├── docs/
│   ├── parameters.md
│   └── m1-pro-16gb-notes.md
└── modelfiles/
    ├── qwen2.5-coder-7b-m1-balanced.Modelfile
    ├── qwen2.5-coder-7b-m1-opencode.Modelfile
    ├── qwen2.5-coder-7b-m1-aider.Modelfile
    ├── qwen2.5-coder-7b-m1-max-context.Modelfile
    ├── deepseek-coder-6.7b-m1-balanced.Modelfile
    ├── deepseek-coder-6.7b-m1-aider.Modelfile
    ├── glm4-9b-m1-balanced.Modelfile
    └── gemma3-12b-m1-careful.Modelfile

Quick start

1) Make scripts executable

chmod +x bin/*.sh bin/ollama-profile

2) Run a profile directly

./bin/ollama-profile-m1-balanced.sh

3) Or use the generic launcher

./bin/ollama-profile m1-stable
./bin/ollama-profile m1-balanced
./bin/ollama-profile m1-opencode
./bin/ollama-profile m1-aider
./bin/ollama-profile m1-max-context
./bin/ollama-profile private
./bin/ollama-profile debug

Profiles

m1-stable

Safest preset for M1 Pro 16 GB when you care more about predictability than throughput.

m1-balanced

Best daily default for local coding on M1 Pro 16 GB.

m1-opencode

More aggressive preset for OpenCode style workflows where the tool may inspect more of the repo and keep slightly more context around.

m1-aider

More aggressive preset for Aider style edit loops where you want fast repeated edits and quick follow-up requests.

m1-max-context

Prioritizes context length over concurrency. Best when repo understanding matters more than throughput.

private

Balanced preset plus cloud-disabled behavior.

debug

Balanced preset with extra logs enabled.

Recommended starting point on M1 Pro 16 GB

For most sessions:

./bin/ollama-profile m1-balanced

For OpenCode:

./bin/ollama-profile m1-opencode

For Aider:

./bin/ollama-profile m1-aider

Build example models

ollama create qwen-coder-m1-balanced -f modelfiles/qwen2.5-coder-7b-m1-balanced.Modelfile
ollama create qwen-coder-m1-opencode -f modelfiles/qwen2.5-coder-7b-m1-opencode.Modelfile
ollama create qwen-coder-m1-aider -f modelfiles/qwen2.5-coder-7b-m1-aider.Modelfile

Design choices

These presets are intentionally opinionated:

• q8_0 KV cache is used as the default balance between memory and quality
• MAX_LOADED_MODELS=1 avoids hidden memory spikes on laptops
• NUM_PARALLEL is kept conservative because raising it together with context can quickly destabilize a 16 GB machine
• KEEP_ALIVE stays long enough to make repeated agent calls feel fast without pinning everything forever

Notes on compatibility

The wrappers are plain shell scripts and should work on both macOS and Linux as long as:

• bash is available
• ollama is on PATH

They only apply settings to the current ollama serve process. They do not edit your shell profile or any system-wide launch settings.