Praex The AI coding agent that answers the question

Praex, an MIT-licensed AI coding agent for the terminal released by praex.ai, is a fork of OpenCode that reads codebases, edits files, runs commands, and keeps resumable sessions, working with any major provider via the user's own API key or with local models through Ollama, LM Studio, llama.cpp and vLLM. Praex adds stricter permission checks for destructive commands, an agent loop that continues when a step stalls, a self-check after headless runs, Remote Control (/rc), persistent memory, and optional hosted models on three tiers — Velox (Free), Faber (Pro) and Lucia (Max) — served on the company's own GPUs. The tool installs via a curl script or prebuilt Linux x64 and macOS binaries with SHA256 checksums, requires Bun, and offers unlimited free usage with a user-supplied key and no Praex account.

The AI coding agent that answers the question. praex.ai https://praex.ai · Download https://praex.ai/download · Chat in the browser https://praex.ai/chat Praex is an open source MIT AI coding agent for the terminal. It reads your codebase, edits files, runs commands, and keeps sessions you can come back to. Use it with your own key for any major provider or with a model running on your own machine. No Praex account is needed for either. curl -fsSL https://praex.ai/install.sh | bash Prebuilt binaries for Linux x64 and macOS Apple Silicon and Intel , with SHA256 checksums, are also on the Releases https://github.com/ailucius/praex-cli/releases page. Set the provider's usual environment variable and start Praex: export ANTHROPIC API KEY=... or OPENAI API KEY, GEMINI API KEY, OPENROUTER API KEY, ... praex Or store a key with praex auth login . Works with OpenAI, Anthropic, Gemini, Groq, DeepSeek, Mistral, OpenRouter, xAI, and any OpenAI-compatible endpoint. praex models lists what is available. Usage with your own key is unlimited and free. Point Praex at Ollama, LM Studio, llama.cpp, vLLM or any other OpenAI-compatible server. Put this in praex.json in your project or in ~/.config/praex/praex.json for every project : { "$schema": "https://praex.ai/config.json", "provider": { "ollama": { "npm": "@ai-sdk/openai-compatible", "name": "Ollama local ", "options": { "baseURL": "http://localhost:11434/v1" }, "models": { "qwen3-coder:30b": { "name": "Qwen3 Coder 30B" } } } } } Praex also runs hosted models we post-train and serve on our own GPUs. Sign in with praex login to use them: | Tier | Model | Plan | |---|---|---| | Velox | fast daily driver | Free | | Faber | heavier coding and reasoning | Pro | | Lucia | frontier-class flagship | Max | No nannying. Praex models answer the question without moralizing, lecturing, or refusing legitimate work. There is a hard floor for genuinely dangerous content, and nothing else. - TUI built for real work: folded tool output, sessions, themes - Permission modes · manual approval by default, loosen when you trust it - Headless runs · praex run "fix the failing test" for scripts and CI, praex serve for a local server - /rc · Remote Control: keep a session running here and continue it from your phone or the web Praex account - Web search on hosted tiers, permission-gated per query - Memory · persistent per-user memory across sessions Requires Bun https://bun.sh . bun install cd packages/opencode bun run script/build.ts --single ./dist/opencode-linux-x64/bin/opencode --version Praex is a fork of OpenCode https://github.com/anomalyco/opencode MIT . We keep its excellent agent core, including support for every provider above, and add our own work on top: stricter permission checks for destructive commands, an agent loop that carries on when a step stalls, a self-check after headless runs, Remote Control /rc , persistent memory, and optional hosted Praex models. Upstream attribution is preserved in LICENSE https://github.com/ailucius/praex-cli/blob/main/LICENSE and THIRD-PARTY-NOTICES.md https://github.com/ailucius/praex-cli/blob/main/THIRD-PARTY-NOTICES.md .