# Bodega One Code: Full Reference

> Bodega One is a developer-tools company. This is the full reference for **Bodega One Code**, the company's IDE product (currently in beta). Sibling products in the family: Bodega One Staff (autonomous AI agent, upcoming), Bodega One HQ (team operations platform, upcoming).

> Bodega One Code is a local-first AI desktop environment for developers.
> Free to use, commercial work included. No subscription. Runs on Windows, macOS, and Linux.
> Full Monaco IDE + AI chat + autonomous coding agent in one app.
> Bring Your Own LLM (BYOLLM): 15+ provider presets, including 9 local providers.
> Air-gap mode: zero network egress, works completely offline.

Last updated: 2026-10-13

---

## What It Is

Bodega One Code is an Electron 40 desktop application combining:
- A full Monaco editor (same engine as VS Code) with syntax highlighting, IntelliSense, and multi-file editing
- An AI chat interface with tool-use capabilities (file read/write, shell, web search, memory)
- An autonomous coding agent that plans, writes, and verifies code in a loop
- A terminal with 3 modes: Shell, Assist (AI suggestions overlay), Agent (full agentic chat)

The key differentiator: it runs entirely on your machine. Your code never leaves your device unless you explicitly configure a cloud LLM provider.

---

## Core Value Propositions

1. **Own It**: Free to use, for personal and commercial work. No tiers, no licensing, no subscription, no renewal.
2. **Private**: Data never leaves your machine. Air-gap mode enforces zero network egress.
3. **Flexible**: Any LLM via BYOLLM. 15+ provider presets: switch in seconds.
4. **Powerful**: Full IDE + AI chat + autonomous coding agent. Not just a chat app.

---

## LLM Providers (15 total)

### Local Providers (9)
- Ollama: most popular local model runner; supports Llama, Mistral, Qwen, DeepSeek, and more
- LM Studio: macOS-first GUI for running GGUF models
- vLLM: high-throughput inference server for GPU clusters
- llama.cpp: low-level C++ inference, CPU and GPU
- LocalAI: OpenAI-compatible local server
- KoboldCpp: creative writing and coding focused local runner
- GPT4All: simple local model runner with built-in model library
- MLX: Apple Silicon optimized inference (macOS only)
- Jan: open-source local AI with a built-in model hub

### Cloud Providers (5)
- OpenAI (GPT-4o, o1, o3)
- Anthropic (Claude Sonnet, Opus, Haiku)
- Groq (fast inference, Llama, Mixtral)
- Together AI (wide model selection)
- OpenRouter (unified API for 100+ models)

### Custom (1)
- Any OpenAI-compatible endpoint: bring your own server URL and API key

---

## Key Features

### Monaco IDE
- Same editor engine as VS Code
- Full syntax highlighting, IntelliSense, multi-cursor, find/replace
- Multi-tab file editing with diff view
- Integrated with the AI agent: agent writes directly into the editor

### AI Chat
- Persistent conversation with full tool-use
- 34+ built-in tools: file_system, glob, grep, shell, web_fetch, web_search, save_memory, scratchpad, str_replace, and more
- Multi-agent support: spawn_worker, delegate_task, get_job_result (gated by model size)
- Semantic memory: 4-layer memory system (session, project, shared, user)

### Autonomous Coding Agent
- Full agentic loop: plan → write → verify → iterate
- Quality Enforcement Layer (QEL): 3-level verification system
  - Level 1: Pattern + compile check after every file write (threshold 70)
  - Level 2: tsc/py_compile proof gate every 2nd write (10s timeout)
  - Level 3: Full verification post-loop (threshold 80 for new files, 50 for modifications)
- SPEC-01 Clarification Interview: agent pauses to ask questions when prompt confidence is low
- Observation masking: ~52% token savings using JetBrains NeurIPS 2025 technique
- KV cache strategy: 40-70% cache reuse via static system prompt, deterministic tool ordering, cache-safe compaction
- Architect/Editor dual-model (in progress): strong model plans, fast model executes

### Air-Gap Mode
- 9 enforcement layers guarantee zero network egress:
  1. Tool filtering: network tools removed from available toolset
  2. Pre-execution guard: second check at execution time
  3. Shell command blocking: curl, wget, npm install, git push blocked
  4. Context assembly guard: external references stripped from system prompt
  5. Auto-updater blocking: no pings to Bodega One servers
  6. UI feedback: visual indicator confirms air-gap is active
  7. Cloud STT blocking: cloud speech-to-text disabled
  8. System prompt filtering: URL patterns stripped
  9. Git IPC blocking: remote git operations blocked
- Shows warning if cloud LLM provider is selected while air-gap is on

### Terminal
- Multi-terminal: multiple simultaneous tabs with independent shells
- 3 modes: Shell (standard PTY), Assist (AI overlay), Agent (full agentic chat)
- OSC 133 shell integration: structured command tracking with exit codes
- WorkflowTemplates: one-click common workflows (install, build, test, git ops)
- Shell security: 13 credential patterns scanned (SSH keys, AWS AKIA*, GitHub PAT ghp_*, OpenAI sk-*, JWT, etc.)

### Themes
- 18 UI palettes: 9 dark + 9 light
- CSS custom property theming: switch without reload

---

## Hardware Recommendations

| VRAM | Recommended Model | Quality |
|------|-------------------|---------|
| <4GB | SmolLM3-3B Q4 | Low |
| 6-8GB | Qwen3-8B Q4_K_M | Good |
| 8-12GB | Qwen3-14B Q4_K_M | Strong |
| 12-16GB | GLM-4.7-Flash 30B-A3B Q4 (MoE) | Excellent |
| 16-24GB | Qwen3.8-27B Q4 | Gold |
| 24-48GB | Llama 3.3 70B Q4 | Frontier |
| Apple 16GB | Qwen3-8B MLX Q4 | Good |
| Apple 64GB+ | Qwen3.8-27B MLX | Gold |

---

## Pricing

**Bodega One Code is free to use, for personal and commercial work.** No paid tiers, no licensing, no machine limits, no subscription, no usage-based billing. Every product feature is included for everyone: full IDE, AI chat, autonomous agent, BYOLLM, air-gap, QEL, Cloud Boost, custom agents, agent scheduling.

Current state: open beta. No waitlist, no password. Download at /download. It is a beta, so expect rough edges and report them on Discord or GitHub.

Your only cost is whichever LLM provider you choose to connect. Local models (Ollama, LM Studio, llama.cpp) cost nothing to run.

---

## Launch Timeline

- April 2026: Waitlist live (retired 2026-06-03)
- May 2026: Beta opens
- June 2026: Free-tier launch - beta opened to everyone, password gate retired
- Full public launch: coming (rolling beta until launch; no fixed date)

---

## Pages

- [Homepage](https://www.bodegaone.ai): Product overview, feature comparison, pricing table, FAQ, newsletter signup
- [Pricing](https://www.bodegaone.ai/pricing): Detailed plan comparison, FAQ, beta offer
- [Documentation](https://www.bodegaone.ai/docs): Quick start guide, core concepts, LLM provider setup
- [Local LLM Guide](https://www.bodegaone.ai/local-llms): Ranked comparison of 37+ local LLMs for software engineering. SWE-bench Verified scores, VRAM requirements, license info, and opinionated picks for every tier from 4 GB to workstation.
- [Changelog](https://www.bodegaone.ai/changelog): Release history from v0.5 to v0.9
- [Blog](https://www.bodegaone.ai/blog): Technical articles on local AI, BYOLLM, and development tools
- [About](https://www.bodegaone.ai/about): Company background and mission
- [Contact](https://www.bodegaone.ai/contact): Contact information and social links
- [Air-Gap Mode](https://www.bodegaone.ai/air-gap): Deep dive on the 9-layer air-gap enforcement system
- [BYOLLM](https://www.bodegaone.ai/byollm): All 15 supported LLM provider presets explained

---

## Blog Posts

- [GitHub Copilot CLI now runs local models](https://www.bodegaone.ai/blog/github-copilot-cli-local-models-offline-mode): Ollama setup, what still reaches GitHub, COPILOT_OFFLINE, Auto routing, and local sandboxing.
- [How to run Qwen3.8-Flash-Next on a gaming PC with Strata](https://www.bodegaone.ai/blog/run-qwen-3-8-flash-next-locally-strata): the 125B MoE on a 12 GB GPU with 32-64 GB RAM; sizes, measured speeds, setup, license caveats.
- [ZCode tried to upload your Git history](https://www.bodegaone.ai/blog/zcode-uploaded-git-history): What Z.ai's coding app packaged and sent, what the toggles missed, and what to do.
- [ChatGPT Pro and Claude Code limits shrank while API prices fell](https://www.bodegaone.ai/blog/chatgpt-pro-claude-code-limits-cut-september-2026): Claude Code weekly limits cut 17% on Sept 14; ChatGPT Pro 200 allowance lowered, Pro 500 added.
- [What Is AGENTS.md? A Guide for AI Coding Agents](https://www.bodegaone.ai/blog/what-is-agents-md): the cross-tool agent instructions file, nested precedence, and which tools read it
- [Why we built a local-first AI IDE](https://www.bodegaone.ai/blog/why-we-built-local-first-ai): The origin story: paying $40/month for tools that sent our code to servers we don't control.
- [The real cost of AI subscriptions (3-year analysis)](https://www.bodegaone.ai/blog/real-cost-of-ai-subscriptions-3-year-analysis): ChatGPT + Claude + Cursor = a number nobody talks about.
- [How QEL works: AI that proves its own code](https://www.bodegaone.ai/blog/how-qel-works-ai-that-proves-its-own-code): The 3-level verification system inside Bodega One Code's agentic loop.
- [Setting up Ollama with Bodega One Code](https://www.bodegaone.ai/blog/setting-up-ollama-with-bodega-one): The setup mistakes that cost people hours, and which models the community actually runs.
- [Air-gap mode: 9 layers that guarantee zero network egress](https://www.bodegaone.ai/blog/air-gap-mode-9-layers-of-enforcement): How each of the 9 enforcement layers works in practice.
- [KV cache: how Bodega One Code gets 40-70% reuse](https://www.bodegaone.ai/blog/kv-cache-how-we-get-40-70-percent-reuse): Three engineering decisions that eliminate redundant LLM computation.
- [Set up an offline AI IDE in 2026](https://www.bodegaone.ai/blog/best-offline-ai-ide-2026): What actually works offline; hardware, setup, and OS-level verification.
- [Which GPU do you actually need for local AI?](https://www.bodegaone.ai/blog/local-ai-gpu-guide): VRAM tier breakdown from 4GB to 48GB with real model recommendations.
- [AI coding tools that work completely offline (2026)](https://www.bodegaone.ai/blog/offline-ai-coding-tools-2026): The complete list of tools that work with zero internet.
- [LM Studio + Bodega One Code: complete setup guide](https://www.bodegaone.ai/blog/lm-studio-bodega-one-setup): Wire them together in under 5 minutes.
- [Air-gapped AI development for regulated industries](https://www.bodegaone.ai/blog/air-gapped-ai-regulated-industries): Finance, healthcare, defense, and legal teams using local AI.
- [How to run DeepSeek locally with Bodega One Code](https://www.bodegaone.ai/blog/deepseek-local-bodega-one): DeepSeek-R1 and V3 on your own hardware, by VRAM tier.
- [GitHub Copilot vs Cursor vs Bodega One Code](https://www.bodegaone.ai/blog/github-copilot-cursor-bodega-one-comparison): Three different approaches to AI-assisted development, compared honestly.
- [Are local LLMs actually good enough for real development work in 2026?](https://www.bodegaone.ai/blog/local-llms-good-enough-2026): The honest answer for most coding tasks: yes.
- [OpenAI is cutting off Cursor](https://www.bodegaone.ai/blog/openai-cutting-off-cursor): OpenAI is ending Cursor's model access Nov 12; the case for owning your AI coding tools with BYOLLM.
- [The harness leaked your CI secrets](https://www.bodegaone.ai/blog/black-hat-coding-agents-ci-secrets): Black Hat 2026 findings on coding agent harness flaws across Claude Code, Gemini CLI, and Codex.
- [DeepSeek is raising API prices](https://www.bodegaone.ai/blog/deepseek-api-price-hike-cheap-cloud-ai): What changed in DeepSeek's API pricing and the local-first alternative.
- [OpenAI's sandbox escape: lessons for coding agents](https://www.bodegaone.ai/blog/openai-sandbox-escape-lessons-for-ai-coding-agents): What a model breaching Hugging Face means for anyone running an autonomous agent.
- [How to run a local LLM for coding](https://www.bodegaone.ai/blog/how-to-run-a-local-llm-for-coding): Pick a runtime, size a model to your GPU, connect an editor, and code offline.
- [Grok Build CLI uploaded private repos](https://www.bodegaone.ai/blog/grok-build-cli-uploaded-private-repos): What xAI's CLI actually sent to its servers, and the local-first fix.
- [Getting started with the Bodega One Code CLI](https://www.bodegaone.ai/blog/getting-started-bodega-one-code-cli): Install the CLI and run your first verified task from the terminal.
- [Google is retiring Gemini CLI](https://www.bodegaone.ai/blog/gemini-cli-shutdown-terminal-agents): Gemini CLI shuts down July 17, replaced by the cloud-only Antigravity CLI.
- [Cursor acquired Continue.dev](https://www.bodegaone.ai/blog/cursor-acquires-continue-dev): Continue.dev was acqui-hired and shut down; the July 15 export deadline and what it means for open source.
- [Continue.dev alternatives after the Cursor deal](https://www.bodegaone.ai/blog/continue-dev-shutdown-alternatives): Where developers who relied on Continue.dev should go next.
- [The verification gap: AI coding agents should prove their code runs](https://www.bodegaone.ai/blog/ai-coding-agent-that-verifies-its-own-code): Why AI coding agents should have to prove their code runs, and how QEL does it.
- [Claude Fable 5 explained](https://www.bodegaone.ai/blog/claude-fable-5-explained): Anthropic's frontier model, its cost, and how to use it without lock-in via BYOLLM.
- [Even Uber hit the AI coding cost wall](https://www.bodegaone.ai/blog/agentic-ai-cost-wall): Uber capped engineer AI spend at $1,500/mo; why metered agentic AI adds up fast.
- [GitHub Copilot, one week in](https://www.bodegaone.ai/blog/github-copilot-week-one-real-bills): Real developer bills after the June 1 AI Credits switch.
- [GitHub Copilot AI Credits decoded](https://www.bodegaone.ai/blog/github-copilot-ai-credits-math-explained): What Pro, Pro+, and Max actually cover, and free BYOK options.
- [GitHub Copilot Business and Enterprise AI credits cut](https://www.bodegaone.ai/blog/github-copilot-business-enterprise-credits-september-2026): Per-user credits fell 3,000 to 1,900 (Business) and 7,000 to 3,900 (Enterprise) on Sept 1, 2026.
- [The IDE-is-dead debate is misdirection](https://www.bodegaone.ai/blog/ides-are-dead-debate-misdirection): Why the "IDEs are dead" soundbite misses the real shift.
- [Best Windsurf alternatives in 2026](https://www.bodegaone.ai/blog/best-windsurf-alternatives-2026): Picking a path after Cognition acquired Windsurf.
- [Best Tabnine alternatives in 2026](https://www.bodegaone.ai/blog/best-tabnine-alternatives-2026): Air-gap-friendly picks after Tabnine killed individual plans.
- [Kilo Pass vs Anthropic API direct](https://www.bodegaone.ai/blog/kilo-pass-vs-anthropic-direct): Cost math comparing Kilo Pass credits to paying Anthropic direct.
- [Roo Code is shutting down: where to go next](https://www.bodegaone.ai/blog/roo-code-shutdown-alternatives): Migration map after Roo Code's May 15, 2026 shutdown.
- [Best GitHub Copilot alternatives in 2026](https://www.bodegaone.ai/blog/best-github-copilot-alternatives-2026): Privacy-first picks after the April 24 Copilot training-data opt-out change.
- [Best Cursor alternatives in 2026](https://www.bodegaone.ai/blog/best-cursor-alternatives-2026): Six honest Cursor alternatives ranked by use case.
- [Air-gapped AI coding in 2026: a developer setup guide](https://www.bodegaone.ai/blog/air-gapped-ai-coding-guide-2026): Developer setup guide for a fully air-gapped AI coding environment.
- [The best Kilo Code alternatives in 2026](https://www.bodegaone.ai/blog/kilo-code-alternative-2026): Alternatives for when Kilo Code's pricing stops being worth it.
- [GitHub Copilot training data opt-out](https://www.bodegaone.ai/blog/github-copilot-training-data-opt-out): What changed when Copilot defaulted to training on your code, and how to opt out.
- [Gemma 4: Google's first Apache 2.0 open model](https://www.bodegaone.ai/blog/gemma-4-apache-open-model-local-ai): Google's first Apache 2.0 open model and why the license change matters.
- [How to use our free VRAM calculator](https://www.bodegaone.ai/blog/how-to-use-vram-calculator): Walkthrough of the free VRAM calculator tool.
- [AI IDE cost comparison calculator](https://www.bodegaone.ai/blog/ai-ide-cost-comparison-calculator): Walkthrough of the free AI IDE cost comparison calculator.
- [How to plan your LLM context window budget](https://www.bodegaone.ai/blog/context-window-planner-for-local-llms): Walkthrough of the free context window budget planner.
- [The real cost of Kilo Code in 2026](https://www.bodegaone.ai/blog/real-cost-of-kilo-code): Investigative pricing breakdown of Kilo Code's actual monthly cost.
- [How to migrate from Tabnine to Bodega One Code](https://www.bodegaone.ai/blog/migrate-from-tabnine-to-bodega-one): Step-by-step migration guide from Tabnine.
- [How to migrate from Windsurf to Bodega One Code](https://www.bodegaone.ai/blog/migrate-from-windsurf-to-bodega-one): Step-by-step migration guide from Windsurf.
- [Augment Code is sunsetting completions: what to do](https://www.bodegaone.ai/blog/augment-code-sunset-completions-what-next): Augment killed completions on non-Enterprise plans; a local-first replacement strategy.

---

## Contact & Community

- Email: hello@bodegaone.ai
- X (Twitter): @BodegaOneAI
- Discord: https://discord.gg/ZWbKVrhEJ7
- GitHub: https://github.com/Bodegaone