Skip to content

camclay/bigo

BranchesTags

Repository files navigation

BigO

Unified Claude + Gemini + Ollama Agent Orchestrator

BigO intelligently routes coding tasks to the most cost-effective AI backend. Simple tasks go to free local Ollama models, while complex work uses Claude's advanced reasoning or Gemini's large context window. Save money without sacrificing quality.

┌─────────────────────────────────────────────────────────────────┐
│                         BigO Conductor                          │
│  ┌───────────┐  ┌──────────────┐  ┌─────────────────────────┐  │
│  │ Classifier │  │ Task Ledger  │  │     Message Bus         │  │
│  │            │  │ (SQLite)     │  │ (coordination)          │  │
│  └───────────┘  └──────────────┘  └─────────────────────────┘  │
└─────────────────────────────────────────────────────────────────┘
                              │
         ┌────────────────────┼────────────────────┐
         ▼                    ▼                    ▼
┌─────────────────┐  ┌─────────────────┐  ┌─────────────────┐
│  Claude/Gemini  │  │   Ollama Tier   │  │  Validator Pool │
│                 │  │                 │  │                 │
│ • Opus/Pro (P0) │  │ • phi3 (fast)   │  │ • Blind review  │
│ • Sonnet (P1-2) │  │ • qwen3 (default│  │ • Multi-model   │
│ • Haiku/Flash   │  │ • Remote server │  │                 │
└─────────────────┘  └─────────────────┘  └─────────────────┘

Features

  • Smart Task Classification: Automatically categorizes tasks by complexity (TRIVIAL → CRITICAL)
  • Cost Optimization: Routes simple tasks to free Ollama, complex ones to Claude or Gemini
  • Remote Ollama Support: Run Ollama on a dedicated GPU server
  • Task Persistence: SQLite ledger for crash recovery and analytics
  • Cost Tracking: See exactly how much you're saving

Quick Start

Prerequisites

  • Go 1.21+
  • Ollama (local or remote)
  • Claude CLI (for Claude backends)
  • Gemini API Key (optional, for Gemini backends)

Installation

Homebrew (macOS/Linux)

brew tap yourusername/tap
brew install bigo

Download Binary

Download from Releases:

# macOS (Apple Silicon)
curl -LO https://github.com/yourusername/bigo/releases/latest/download/bigo-darwin-arm64
chmod +x bigo-darwin-arm64
sudo mv bigo-darwin-arm64 /usr/local/bin/bigo

# macOS (Intel)
curl -LO https://github.com/yourusername/bigo/releases/latest/download/bigo-darwin-amd64
chmod +x bigo-darwin-amd64
sudo mv bigo-darwin-amd64 /usr/local/bin/bigo

# Linux (amd64)
curl -LO https://github.com/yourusername/bigo/releases/latest/download/bigo-linux-amd64
chmod +x bigo-linux-amd64
sudo mv bigo-linux-amd64 /usr/local/bin/bigo

Build from Source

git clone https://github.com/yourusername/bigo.git
cd bigo
go build ./cmd/bigo
sudo mv bigo /usr/local/bin/

Initialize a Project

cd your-project
bigo init

This creates a .bigo/ directory with:

  • ledger.db - SQLite database for task tracking
  • config.yaml - Configuration file

Run Your First Task

# Dry run (classify without executing)
bigo run -n "fix the typo in README.md"

# Execute
bigo run "fix the typo in README.md"

# Check status
bigo status

Architecture

Two-Machine Setup

BigO is designed for a common setup where you have:

  1. Primary Machine (laptop/workstation): Runs the bigo CLI and Claude
  2. GPU Server: Runs Ollama with larger models
┌──────────────────────┐         ┌──────────────────────┐
│   Primary Machine    │         │     GPU Server       │
│                      │         │                      │
│  ┌────────────────┐  │  HTTP   │  ┌────────────────┐  │
│  │   bigo CLI     │──┼─────────┼─▶│    Ollama      │  │
│  └────────────────┘  │  :11434 │  │                │  │
│          │           │         │  │  • phi3        │  │
│          ▼           │         │  │  • qwen3       │  │
│  ┌────────────────┐  │         │  │  • deepseek    │  │
│  │  Claude/Gemini │  │         │  └────────────────┘  │
│  └────────────────┘  │         │                      │
└──────────────────────┘         └──────────────────────┘

Task Classification

Tier Complexity Backend Use Case
T0 TRIVIAL Ollama (fast) Typos, formatting, comments
T1 SIMPLE Ollama (default) Add function, fix obvious bug
T2 STANDARD Claude Sonnet / Gemini Flash New feature, refactoring
T3 COMPLEX Claude Sonnet / Gemini Pro Architecture, multi-file
T4 CRITICAL Claude Opus / Gemini Pro Security, auth, payments

Configuration

Primary Machine

Edit .bigo/config.yaml:

conductor:
  classifier_model: claude:sonnet
  max_retries: 3
  validation_timeout: 300s

workers:
  claude:
    enabled: true
    max_concurrent: 2
    models:
      opus: claude-opus-4-5-20251101
      sonnet: claude-sonnet-4-20250514
      haiku: claude-haiku-3-5-20241022
    cost_limits:
      daily_usd: 50.0
      per_task_usd: 5.0

  gemini:
    enabled: true
    api_key: "YOUR_GEMINI_API_KEY"
    models:
      flash: gemini-1.5-flash
      pro: gemini-1.5-pro

  ollama:
    enabled: true
    endpoint: http://your-gpu-server:11434  # Remote Ollama
    max_concurrent: 4
    models:
      fast: phi3:mini-16k      # 3.8B - trivial tasks
      default: qwen3:8b        # 8.2B - simple tasks
      reasoning: qwen3:8b-8k   # Extended context

validators:
  pool_size: 5
  timeout: 120s
  backends:
    - claude:sonnet
    - ollama:qwen3:8b

ledger:
  path: .bigo/ledger.db

GPU Server (Ollama)

See docs/ollama-server-setup.md for detailed setup.

Quick setup:

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

# Pull models
ollama pull phi3:mini-16k
ollama pull qwen3:8b

# Configure for network access
sudo systemctl edit ollama.service
# Add: Environment="OLLAMA_HOST=0.0.0.0"

sudo systemctl restart ollama

# Verify
curl http://localhost:11434/api/tags

Commands

bigo init              # Initialize in current directory
bigo run "task"        # Execute a task
bigo run -n "task"     # Dry run (classify only)
bigo classify "task"   # Test classifier
bigo status            # View stats and cost savings
bigo config            # View configuration

Cost Savings Example

BigO Status
═══════════════════════════════════════
Tasks:      47 total (2 pending, 45 completed)
Executions: 52 total
───────────────────────────────────────
Cost Breakdown:
  Claude/Gemini: $1.2340 (12 tasks)
  Ollama:        $0.0000 (35 tasks)
  Savings:       $1.7500 (58.6%)
═══════════════════════════════════════

Roadmap

  • Validation System: Blind validators that review worker output
  • Parallel Workers: Multiple concurrent Ollama instances
  • Kubernetes Support: Scale Ollama across a cluster
  • OpenCode Integration: Use OpenCode for tool-enabled local execution
  • Web Dashboard: Visual task management and analytics

Project Structure

bigo/
├── cmd/bigo/              # CLI entry point
├── internal/
│   ├── cli/               # Command implementations
│   ├── conductor/         # Orchestrator and classifier
│   ├── config/            # Configuration management
│   ├── ledger/            # SQLite state management
│   ├── workers/           # Ollama and Claude workers
│   ├── validators/        # Validation system (planned)
│   └── bus/               # Message bus (planned)
├── pkg/types/             # Shared types
├── docs/                  # Documentation
├── examples/              # Example configurations
└── scripts/               # Setup and utility scripts

Inspiration

BigO draws inspiration from:

License

MIT License - see LICENSE for details.

Contributing

Contributions welcome! See CONTRIBUTING.md for guidelines.

About

No description, website, or topics provided.

Resources

Readme

License

MIT license

Contributing

Contributing
Activity

Stars

2 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages