Skip to main content

Everything you need to get started.

Bodega One is built to be picked up fast. Install it, connect your LLM, and you're building with AI in minutes. No config files to wrestle with.

Up and running in three steps.

No Docker. No environment setup. No CLI.

  1. Download and install

    Grab the installer for your platform: Windows, macOS, or Linux. Double-click, follow the prompts. No PATH setup, no CLI required.

  2. Connect your LLM

    Open Settings → LLM Providers. Pick a preset (Ollama for local, OpenAI or Anthropic for cloud) and paste your API key or endpoint URL. Takes about 30 seconds.

  3. Start building

    Open a folder in Code Mode or start a conversation in Chat Mode. The AI has full access to your files and 23 built-in tools from the first message.

Beta is live now. Join the waitlist to be first in line.

Watch it in action.

Two short walkthroughs covering the most common ways to bring a local LLM into Bodega One.

Ollama in under a minute

Already running Ollama? Bodega One auto-detects it. New to Ollama? We walk through the install, pulling your first model, and using it in Code mode.

Local llama.cpp, no CLI

Managed runtime, GGUF catalog, hot-swap, crash recovery. Click a model and it loads. Pull anything from Hugging Face without touching a config file.

More walkthroughs on the Bodega One YouTube channel.

Four things to understand.

These aren't marketing terms. They're the actual architecture of how Bodega One works.

Code Mode

Full IDE with an AI agent

Monaco editor, file tree, multi-terminal, and an autonomous coding agent in one window. The agent writes real diffs, not suggestions. You review and apply.

Chat Mode

Conversational AI with real tools

Full-screen AI chat with persistent memory and 23 built-in tools. The AI can read files, run shell commands, search the web, and more. All from the conversation.

QEL

Quality Enforcement Layer

Every code change the agent writes passes through 5 verification stages before you see it: contract extraction, incremental checks, proof gates (tsc, pytest, py_compile), and targeted line-level repair. The AI can't game its own checks.

How QEL works →

BYOLLM

Bring Your Own LLM

Connect any of 10+ supported providers. Run Ollama locally for full privacy. Switch to Claude for complex reasoning. Swap models any time. Never locked in.

What BYOLLM means →

Connect any model you want.

15 provider presets built in. Open Settings → LLM Providers, pick a preset, enter your key or local endpoint. Takes about 30 seconds.

Local: runs on your machine

Ollamarecommended

Best for privacy: runs fully local

LM Studio

Local models with a GUI

vLLM

High-throughput local serving

llama.cpp

Lightweight GGUF inference

LocalAI

OpenAI-compatible local API

KoboldCpp

Flexible GGUF serving

GPT4All

Desktop local models

MLX

Optimized for Apple Silicon

Jan

Local AI desktop client

Cloud: bring your own key

OpenAI

GPT-4o, o3, and the o-series

Groq

Fast inference for open models

Together AI

Open models at scale

OpenRouter

Multi-provider gateway

Azure OpenAI

Enterprise OpenAI deployment

Custom endpoint

Any OpenAI-compatible API

Using Ollama? That's the fastest path to full privacy.

Install Ollama, pull a model (ollama pull llama3.2), then set the endpoint to http://localhost:11434 in Bodega One. Nothing leaves your machine.

Pick a model for your hardware.

Not sure what to run? Match your GPU to the right model. These are tested recommendations for agentic coding workloads in Bodega One.

Your GPU / RAMRecommended modelQuality
< 4 GBSmolLM3-3B Q4Basic
6–8 GBQwen3.5-9B Q4_K_MGood
8–12 GBQwen3.5-27B Q4Strong
12–16 GBGemma 4 26B MoE Q4Excellent
16–24 GBQwen3.6-27B Q4Gold
24–32 GBQwen3.6-35B-A3B Q4Gold
48 GB+GLM-5.1 / Llama 4 ScoutFrontier
Apple Silicon 16 GBQwen3.5-9B MLX Q4Good
Apple Silicon 64 GB+Qwen3.6-27B MLXGold
Apple Silicon 128 GB+GLM-5.1 MLX / Qwen3.6-35B MLXFrontier

Not sure how much VRAM you have? Run nvidia-smi on Windows/Linux or check System Report → Graphics on Mac. VRAM calculator →

Still have questions?

Discord is the fastest way to get answers from the team and other beta users. Or join the waitlist and we'll walk you through setup on day one.