Vexi reads your project, edits your files, and remembers your decisions. Bring your own key from any provider. No signup, no server, no cost.
Built for developers who live in the terminal.
npm install -g vexi-cli and you're running. Works on Windows, Mac, and Linux. No config files.
Auto-detects your provider from the key. Groq, Gemini, Anthropic, OpenAI, OpenRouter. Stored locally only.
Context Compression Engine keeps a running summary of your decisions. No more re-explaining your project.
Scans your entire codebase, not just the open file. Respects .gitignore, skips node_modules.
Export any session as an interactive HTML you can share. Watch the AI build, step by step.
The first AI tool that explains any code in Arabic, Spanish, Portuguese, French, or English.
Runs pip install, gcc, javac, cargo and more directly from chat. Vexi auto-executes build commands for Python, Java, C, Rust, Go, and any language.
Mines your past sessions for corrections and builds a personal skill file — injected into every future session so you stop repeating yourself.
Stage, draft an AI commit message, and push — all from the chat. Auth pre-flight checks credentials before every push so you're never blocked mid-flow.
Paste any OpenAI-compatible endpoint URL and Vexi identifies the provider, fetches the model list, and saves your config — no key-prefix guessing needed.
Every AI file edit is snapshot before it happens. Type /undo to roll back, /redo to reapply, /history to browse the full change log. Safe to experiment.
vexi update runs the global npm update in one command. Vexi checks for a newer version daily and shows a notice at startup. Uninstall cleanly with vexi uninstall.
vexi -p "your prompt" runs a single non-interactive reply — perfect for scripts and CI pipelines. Add --yes to auto-run any commands it proposes.
Structured read/write/edit tools replace fragile shell patching — exact substring edits, precise per-file undo snapshots, and native function-calling on providers that support it.
See real token usage per session and an estimated dollar cost — anytime with /usage and a summary at exit. Free-tier models read as $0. You always know what you're spending.
| Feature | Vexi | OpenCode | Claude Code | Cursor |
|---|---|---|---|---|
| One-command install | ✓ | ✓ | ✓ | app |
| BYOK any provider | ✓ | ✓ | limited | partial |
| Context memory | ✓ | partial | partial | ✓ |
| Multi-language explain | ✓ 5 languages | via model | via model | via model |
| Open source | ✓ MIT | ✓ | ✕ | ✕ |
| No server needed | ✓ | ✓ | ✕ | ✕ |
| Builds any language | ✓ | ✓ | ✓ | ✓ |
| Git push from chat | ✓ | ✕ | ✓ | partial |
| URL endpoint setup | ✓ | ✕ | ✕ | ✕ |
| Token & cost tracking | ✓ | ✕ | ✓ | partial |
Auto-detected from the key format, or use vexi setup to paste any OpenAI-compatible endpoint URL.
Most AI coding tools are complex to install, locked to one provider, or send your code to a server. Vexi takes the opposite path.
It's a single npm install, works with any API key you already have, and runs 100% on your machine. Your code and keys never leave your computer.
Vexi is open source under the MIT license and built for developers everywhere — with first-class support for Arabic, Spanish, Portuguese, French, and English.
Questions, feedback, bug reports, or partnership ideas — reach out anytime.
hello@vexi.proVexi runs entirely on your machine. The Vexi CLI does not collect, transmit, or store any of your data on our servers — because there are no Vexi servers.
Vexi is free and open-source software provided under the MIT License.
Free, open source, and yours to keep. One command away.