LocalAIbundle

LocalAIbundle installs and maintains a fully local AI coding stack for macOS Apple Silicon. It sets up Ollama, coding models, VS Code, Continue.dev, diagnostics, repair tooling, and offline bundle support with no cloud accounts or API keys.

Your code stays on your machine.

What It Provides

Capability	Details
Local inference	Ollama on localhost:11434 with Apple Metal acceleration
IDE integration	VS Code + Continue.dev for chat, autocomplete, inline edits, and codebase context
Hardware-aware profiles	Automatically selects models by RAM, or accepts explicit profile/model overrides
Safe configuration	Backs up existing Continue config and VS Code settings before writing
Diagnostics	doctor checks Ollama, models, LaunchAgent, VS Code, Continue, config, and telemetry
Repair mode	repair fixes common stale or missing setup pieces
Offline workflow	bundle packages the repo and local Ollama model cache for air-gapped installs
Config validation	validate-config checks the generated Continue YAML shape and required roles
Install reports	Writes JSON reports under ~/.localaibundle/ for auditing and troubleshooting
Support bundles	issue-report creates a redacted archive for troubleshooting
CI coverage	Bash syntax, ShellCheck, whitespace checks, and unit tests run in GitHub Actions

Quick Start

Prerequisites

• macOS on Apple Silicon (M1/M2/M3/M4)
• 16GB RAM minimum, 32GB+ recommended
• 20GB+ free disk space for the default profile
• Internet connection for the first model download

Install

git clone https://github.com/DevenDucommun/LocalAIbundle.git
cd LocalAIbundle
chmod +x install.sh

# Preview without making changes
./install.sh --dry-run

# Install the default local coding stack
./install.sh install

Package-manager installs expose the same interface as localaibundle. From a clone, you can use bin/localaibundle as that stable command wrapper.

Verify

./install.sh doctor
./install.sh doctor --json
./install.sh test
./install.sh self-test --no-network

Model Profiles

The default auto profile selects models by RAM. You can also choose a profile explicitly.

Profile	Completion	Chat/Edit	Best For
auto	RAM-based	RAM-based	Most users
fast	qwen2.5-coder:1.5b	qwen2.5-coder:3b	Lightweight laptops
balanced	qwen2.5-coder:1.5b	qwen2.5-coder:7b	16GB machines
professional	qwen2.5-coder:3b	qwen2.5-coder:14b	Daily coding on 24GB+
agentic	qwen2.5-coder:3b	qwen3-coder:30b	Larger coding tasks and long-context work
max	qwen2.5-coder:7b	qwen2.5-coder:32b	High-memory machines

Automatic RAM tiers:

RAM	Completion	Chat/Edit
16-23GB	qwen2.5-coder:1.5b	qwen2.5-coder:7b
24-47GB	qwen2.5-coder:3b	qwen2.5-coder:14b
48GB+	qwen2.5-coder:7b	qwen2.5-coder:32b

Example overrides:

./install.sh install --profile agentic
./install.sh install --chat-model qwen3-coder:30b --completion-model qwen2.5-coder:3b

Commands

./install.sh install        # Install and configure the full stack
./install.sh doctor         # Diagnose local setup without changing anything
./install.sh repair         # Fix common setup/configuration issues
./install.sh status         # Print component status
./install.sh status --json  # Print machine-readable status
./install.sh test           # Run inference smoke tests
./install.sh self-test      # Run non-mutating installer/release checks
./install.sh issue-report   # Create a redacted troubleshooting archive
./install.sh validate-config # Validate Continue config for the selected profile
./install.sh bundle         # Create an offline bundle
./install.sh uninstall      # Remove LocalAIbundle components
./install.sh --help         # Show all commands and flags

Useful install modes:

./install.sh install --models-only
./install.sh install --config-only
./install.sh install --no-vscode
./install.sh install --no-continue
./install.sh install --no-launchagent
./install.sh install --no-model-pull
./install.sh uninstall --preserve-models
./install.sh uninstall --preserve-config
./install.sh issue-report --issue-output localai-report.tgz

Offline Installs

Create a bundle on a machine that already has the desired Ollama models:

./install.sh bundle --profile professional --output LocalAIbundle-offline-professional.tar.gz

Move the tarball to the offline Mac, then run:

./install.sh install --offline LocalAIbundle-offline-professional.tar.gz

The offline bundle includes the installer, docs, manifest, and local Ollama model cache when ~/.ollama/models exists.

Documentation

• FAQ
• Privacy and security model
• Troubleshooting
• Model profile guide
• Manual setup comparison
• Packaging
• macOS QA checklist

How It Works

VS Code + Continue.dev
  |  chat, edit, autocomplete, @codebase
  v
Ollama on localhost:11434
  |  local model files in ~/.ollama
  v
Apple Silicon / Metal GPU

Generated files:

Path	Purpose
~/.continue/config.yaml	Continue.dev local model configuration
~/Library/LaunchAgents/com.localai.ollama.plist	Ollama auto-start on login
~/.localaibundle/install-report-*.json	Install diagnostics and selected model report
~/.continue/config.yaml.bak.*	Timestamped backup of previous Continue config
LocalAIbundle-issue-report-*.tar.gz	Redacted local troubleshooting archive, only when requested

Privacy

• Ollama inference runs locally.
• Continue is configured to use local Ollama models.
• Continue telemetry is disabled in VS Code settings.
• No API keys or cloud accounts are required.
• After the first install or an offline bundle import, the stack can run without internet.

Development

Run the same checks as CI:

bash -n install.sh
shellcheck install.sh bin/localaibundle tests/run.sh tests/install-sandbox.sh scripts/package-release.sh scripts/package-pkg.sh scripts/package-dmg.sh scripts/notarize-artifact.sh scripts/notarize-pkg.sh scripts/demo.sh scripts/docker-test.sh
python3 -m py_compile scripts/*.py
bash tests/run.sh
bash tests/install-sandbox.sh
bash scripts/package-release.sh
git diff --check

The unit test harness uses LOCALAIBUNDLE_SOURCE_ONLY=true plus LOCALAIBUNDLE_TEST_* hardware overrides so model selection and config generation can be tested on non-macOS CI runners without touching a real installation.

The sandbox install test creates a temporary HOME and fake ollama, code, curl, brew, and launchctl commands. It runs a non-dry-run install, doctor, and inference smoke test without modifying the developer machine.

Create release artifacts:

bash scripts/package-release.sh

This writes dist/LocalAIbundle-<version>.tar.gz, a matching .sha256 checksum, and a release manifest.

Package scaffolding:

# Developer-friendly package-manager path
cp packaging/homebrew/localaibundle.rb /path/to/homebrew-localai/Formula/

# Native macOS package, unsigned unless SIGN_IDENTITY is set
bash scripts/package-pkg.sh

# DMG containing the package installer, unsigned until notarized
bash scripts/package-dmg.sh

Homebrew tap instructions should be published only after the tap exists and its formula passes brew test.

Docker can run Linux-portable checks:

docker build -f Dockerfile.test -t localaibundle-test .
docker run --rm localaibundle-test

Docker does not validate macOS LaunchAgents, VS Code app layout, Apple Silicon detection, signing, or notarization.

Record a terminal demo:

bash scripts/demo.sh

Manual macOS QA steps live in docs/macos-qa.md.

Uninstall

./install.sh uninstall

Removes the user-local Ollama app, downloaded models, LocalAIbundle LaunchAgent, Continue.dev extension, and Continue config. It does not remove VS Code itself.