Cut your AI bill up to 90%

AI tools charge you by the word. OIOXO makes your AI bill up to 90% smaller — and it works inside the tools you already pay for: GitHub Copilot, Claude, Cursor, Windsurf, Gemini, Codex. One small install, the same answers, a fraction of the cost.

It does this by sending the AI only the small slice of your project it actually needs to answer — instead of dumping in whole files. A question that would cost ~50,000 words comes back as ~5,000. OIOXO also brings a coding assistant that runs right on your machine, so the code you write and the questions you ask never leave your device.

Private coding on your own machine

• The AI runs on your machine. OIOXO picks a coder that fits your computer (~0.3–4.5 GB, matched to its real memory and graphics) and downloads it once. From then on, your code, your questions and the answers never leave the device.
• It reads your real project. Every answer is built from the actual files that matter — and each reply shows you which files it used, and how many words that saved you.
• Built for real code. Answers stream in, code blocks come with Copy / Insert at cursor / New file, right-click Explain / Improve /Add selection, and Ctrl+Alt+O opens the assistant panel.
• Private image understanding. Attach a screenshot — an error message, a UI mockup, a diagram — and OIOXO reads it on your device to help with the answer. Your screenshots never leave the machine; even with your own key, only the text it pulls out is ever sent.
• A bigger model when you want one. Bring your own key from OpenAI, Claude, Gemini, Groq, Mistral, xAI, DeepSeek, OpenRouter, a local Ollama, or any OpenAI-compatible service — your key stays in VS Code, and calls go straight to the provider. OIOXO trims the bill on every one of them.

Code in your terminal — oioxo code

OIOXO isn't only there to make other tools leaner — it's a coding helper in its own right, on the command line. oioxo code "<task>" reads your project, plans an edit, and runs it on your machine (your tests or type check) before showing you the change and asking to apply it.

• Checked, not guessed. Every change is run through your tests on a throwaway copy first; if it fails, OIOXO fixes it and tries again. Your real files are only written after you approve the change.
• Private and on your device. Run oioxo code --local for the built-in coder that runs on your own machine — no key, nothing uploaded (one-time npm i -g node-llama-cpp). Or point it at a local Ollama model.
• Or bring any key. OpenAI, Anthropic, Groq, Mistral, Together, a local Ollama, or any OpenAI-compatible service — and OIOXO still hands even a big model ~10× fewer words to read.

# fully local with Ollama — private, no key, no tokens
OIOXO_PROVIDER=ollama OIOXO_MODEL=your-model oioxo code "add input validation"

# with your own key (~10x fewer words to read -> ~10x cheaper)
OPENAI_API_KEY=...  oioxo code "fix the failing test" --verify "npm test"

Pin a provider once instead of repeating env vars — oioxo model use anthropic and oioxo config set apiKey <key> — then just oioxo code "…". Check anytime with oioxo doctor.

Why your bill drops by up to 90%

AI tools normally read whole files to answer a question. OIOXO hands them only the small slice that matters instead.

Install — pick one

• VS Code: download the .vsix and run code --install-extension oioxo-vscode-0.12.11.vsix (works in VS Code, Cursor, Windsurf & VSCodium). That's the whole setup — when you open a folder, OIOXO connects Copilot, Claude and Cursor for you and shows what you're saving, live, in the status bar.
• Terminal (any editor): npm i -g oioxo-mcp, then oioxo-mcp login and oioxo-mcp init in your project. init detects your agents and writes their configs — merged, never overwritten.
• Other VS Code-family editors (VSCodium, code-server, and editors that side-load extensions): download the .vsix and install it via Extensions → ⋯ → Install from VSIX.

What your AI tool gains

• get_context — the exact code in play, plus the shape of the other files it depends on. Pulled in automatically before your tool starts reading files.
• get_impact — everything that would be affected by changing a file, before a rename or rewrite breaks something.
• get_skeleton — the outline of a file at roughly a tenth of the words.
• remember / recall — lasting notes about your project, kept in .oioxo/ and shared with the OIOXO app, so your tool stops re-learning the project every time.

Why it makes answers better, not just cheaper

An AI buried in 75,000 words of barely-related files misses things. OIOXO keeps the useful parts and drops the noise: the code in play arrives in full, the rest as a short outline. Less to read, less to get lost in, sharper edits.

Your code stays private

OIOXO reads and searches your project entirely on your own machine — your code is never uploaded. The only thing OIOXO's servers ever see is the number of words you saved, for your meter. Same rule as everything else here: privacy by design.

Plans

Free accounts include a generous monthly savings allowance — enough to feel the difference on real work. OIOXO Pro ($3.99/mo) makes the savings unlimited, across every tool, every project, every machine. The status bar (and oioxo-mcp status) always shows where you stand.