Goose Local Agents - AdvanceCyber.ai

See It In Action

Watch how Goose Local brings offline AI coding capabilities to your workflow

Coding Unchained - Offline AI Development

Why Go Local?

100% Offline

Works on planes, in tunnels, anywhere without WiFi

Privacy First

All processing happens locally - nothing leaves your machine

Free Forever

No API costs, no subscriptions, no token limits

Same Agents

Same agent library you know, converted for local use

LocalCLI: Claude Code Interface

A Python CLI that replicates Claude Code's UX while running entirely offline with Goose + Ollama

Familiar Interface

Same commands, same approval workflow, same output formatting you know from Claude Code.

• /help, /clear, /compact commands
• Tool approval with y/n/e/a/? keys
• Unified diff display with syntax highlighting
• Markdown rendering in responses

Context Translation

Seamlessly switch between Claude Code and Goose with automatic context translation.

• Bidirectional CLAUDE.md ↔ .goosehints
• Tool reference mapping (MCP → Goose extensions)
• Section header translation
• Git hook integration for auto-sync

Install LocalCLI

# Clone and install
git clone https://github.com/advancecyber/localcli.git
cd localcli
pip install -e .

# Run LocalCLI
localcli

Setup Guide

1. Install Goose

# macOS
brew install goose

# Linux / WSL
curl -fsSL https://github.com/block/goose/releases/latest/download/install.sh | bash

2. Install Ollama

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Start Ollama service
ollama serve

3. Pull Local Models

# Recommended: Qwen3 Coder 30B (best balance)
ollama pull qwen3-coder:30b

# Heavy option: Llama 3.3 70B (needs 48GB+ VRAM)
ollama pull llama3.3:70b

4. Configure Goose

# Run configuration wizard
goose configure

# Select: Configure Providers → Ollama
# Enter OLLAMA_HOST: http://localhost:11434
# Select model: qwen3-coder:30b

5. Install Extensions

# Download and extract
wget https://advancecyber.ai/goose-extensions.zip
unzip goose-extensions.zip
cd goose-extensions

# Run setup
./setup.sh

6. How to Use LocalCLI

# Start interactive session (Claude Code-style REPL)
localcli

# Execute a single command and exit
localcli -e "refactor the auth module"

# Use a different model
localcli -m llama3.3:70b

# Resume a previous session
localcli -r abc123

# Inside the REPL - familiar commands
> /help          # Show available commands
> /clear         # Clear conversation
> /compact       # Summarize context

74 Recipes Included

All public agents and skills converted to Goose recipe format

Development Agents

• architect - System design
• auto-code - Implementation
• refactor - Code improvement
• bug-find - Debugging
• code-reviewer - Quality checks

Orchestration Agents

• conductor - Workflow orchestration
• pm - Project management
• critic - Quality validation
• project-setup - Initialization

Specialized Agents

• ciso - Security review
• qa-guy - Testing
• doc-gen - Documentation
• api-docs - API specs
• n8n - Workflow automation

Claude Code vs Goose

Feature	Claude Code	Goose Local
Model Quality	Claude Sonnet/Opus (state-of-art)	Qwen Coder (~90% of Claude)
Connectivity	Requires internet	Works fully offline
Cost	API usage fees	Free (your hardware)
Privacy	Data sent to Anthropic	100% local
Context Window	200K tokens	32K-128K (model dependent)

Hardware Requirements

High-End

llama3.3:70b

• 48GB+ VRAM or
• 64GB RAM (unified memory)
• ~8 tok/s on RTX 4090

Tip: Apple Silicon Macs with unified memory work great. The qwen3-coder:30b model runs well on M3 Max with 64GB.

Ready to Go Local?

Download the extensions package and start coding offline in minutes.

Download goose-extensions.zip

74 recipes • Setup script included • Works offline

System Architecture

The Offline Code Twin architecture - complete AI coding capabilities without internet connectivity

Click to view the full PDF documentation