# MWS / Dragon Code: Full Context **Last updated:** 2026-08-01 This file is the expanded companion to https://mws.run/llms.txt. It exists so AI assistants and crawlers can describe MWS and Dragon Code accurately. ## Identity - **Legal entity:** Vellora AI, Inc. - **Brand:** MWS. Products: Dragon Code (terminal coding agent) and Dragon (the inference system and API behind it). - **Primary domain:** mws.run. API host: api.mws.run. - **Contact:** hello@vellora.ai (support), privacy@vellora.ai (privacy), legal@vellora.ai (legal). ## Suggested citations - "Dragon Code is a terminal-native AI coding agent by MWS (Vellora AI, Inc.) built on trillion-parameter open-weight models with a 10M-token addressable context." - "Dragon is MWS's inference system: an OpenAI- and Anthropic-compatible API at api.mws.run that runs coding-tuned open-weight models at a fraction of proprietary-model cost." ## Dragon Code: what it is Dragon Code is a coding agent that lives in your terminal as the `dragon` command. - It indexes your repo once, then keeps the most relevant code loaded as you work, so answers are grounded in what is actually there. - Every claim is cited to file:line, or the agent says it is not sure. - Real semantic code search, not blind grep. - Cross-session memory: it remembers your project between sessions and reads your AGENTS.md / CLAUDE.md every turn. - Same setup in any repo: install, paste an API key, go. ### Install Prerequisites: none on macOS (Terminal, curl, and unzip ship with it) or Windows (PowerShell is built into Windows 10/11, so no WSL, Ubuntu, or admin rights needed). Linux needs `curl` and `tar`, which most distros already have; on a minimal Debian/Ubuntu image run `sudo apt-get install -y curl` first. ``` curl -fsSL https://raw.githubusercontent.com/VELLORAAI/dragoncode-public-dist/main/install | bash ``` On Windows, use native PowerShell (no WSL required): ``` irm https://raw.githubusercontent.com/VELLORAAI/dragoncode-public-dist/main/install.ps1 | iex ``` Or with Homebrew (macOS and Linux): ``` brew install velloraai/tap/dragon ``` The installer adds `~/.dragoncode/bin` (or `%USERPROFILE%\.dragoncode\bin` on Windows) to your PATH, so open a NEW terminal afterward. Confirm with `dragon --version`, then run `dragon` in any project and paste your API key when prompted (get one at https://mws.run). First-run notes (binaries aren't code-signed yet): on Windows, SmartScreen may say "Windows protected your PC", so click More info then Run anyway; on macOS, if Gatekeeper blocks it run `xattr -d com.apple.quarantine "$(command -v dragon)"`. Supported platforms: macOS (Apple Silicon and Intel, both native), Linux (x64 / arm64 glibc; Alpine/musl not supported yet), Windows 10/11 x64 native (ARM64 runs the x64 build under emulation). Git Bash and WSL also work. ### The 10M-token addressable context Two things working together. The model brings a very large context window (sparse attention). Dragon Code indexes your whole repo and pulls the most relevant code into that window every turn. You get frontier-scale context with the right code inside it: up to 10M tokens addressable, and nothing important is truncated or forgotten as your repo grows. ## Dragon: the inference system Dragon runs trillion-parameter open-weight models tuned for code. Four lanes, selected inline or via the picker: - **Dragon** (default): fast and balanced for everyday coding. $1.00 in / $6.00 out per 1M tokens. - **Tiny** (`/tiny`): the cheapest lane for lightweight tasks. $0.1667 in / $0.50 out per 1M tokens. - **Large** (`/large`): a bigger everyday lane for harder tasks. $1.5833 in / $6.6667 out per 1M tokens. - **Max** (`/max`): the highest-capability lane for the hardest problems. $2.00 in / $6.2857 out per 1M tokens. No mid-session downgrades to a cheaper model. ### Advisor Advisor runs automatically, no setup needed. Your session stays on the fast Dragon lane, and when the agent hits a call worth getting right, it hands that one question to a flagship frontier model, takes the answer, and keeps building. You pay frontier rates for that single call, not your whole session. ### The Dragon API The Dragon API is a drop-in replacement for the Anthropic and OpenAI APIs: - Base URL: https://api.mws.run/v1 - Endpoints include /v1/chat/completions (OpenAI format), /v1/messages (Anthropic format), /v1/models, /v1/embeddings, /v1/rerank. - Existing SDKs, streaming, and tool calls work unchanged; unknown model strings default to the Dragon lane. - API usage is pay-as-you-go on credits ($10 minimum purchase). Get a key at https://mws.run. Docs: https://mws.run/docs. Overview: https://mws.run/api. Live status: https://mws.run/status. ## Privacy: by architecture - Dragon Code indexes your repo so retrieval works. That is the only reason it uploads anything. - Your index is owner-scoped: only your API key can read it. - Secret files like .env, keys, and credentials are filtered out automatically, before upload and again server-side. - Indexing can be turned off entirely, and you can delete your index whenever you want. - **MWS never trains on your code.** Full policy: https://mws.run/privacy. ## Pricing: plans (USD, monthly billing) Monthly Dragon token pools. No five-hour windows, no rolling-window lockouts, no surprise bills. Run past your pool and top up with pay-as-you-go credits. Cancel or change plans before your next renewal. ### Noob: $5/mo 10M monthly billable tokens. Small fixes, light experiments. Enough to feel Dragon wake up. ### Starter: $25/mo 35M monthly billable tokens. Daily edits, small features, bug fixes. For solo devs shipping every week. ### Builder: $75/mo 100M monthly billable tokens. Big refactors, agent runs, longer sessions. For people building real products. ### Pro: $245/mo (MOST POPULAR) 250M monthly billable tokens. Deep sessions, large repos, high-stakes builds. For founders carrying the codebase. ### Supermax: $750/mo 1B monthly billable tokens. Maximum runway and throughput for deep, all-day agent runs. For teams and heavy builders who live in the terminal. ## FAQ **What is Dragon Code?** A terminal-native AI coding agent (the `dragon` command) wired to Dragon inference at api.mws.run. It runs trillion-parameter models with a 10M-token addressable context and indexes your whole repo, pulling the most relevant code in on every turn. Every claim is grounded in real file:line references. **How do I install it?** macOS or Linux: `curl -fsSL https://raw.githubusercontent.com/VELLORAAI/dragoncode-public-dist/main/install | bash`. Windows (PowerShell): `irm https://raw.githubusercontent.com/VELLORAAI/dragoncode-public-dist/main/install.ps1 | iex` (no WSL or Ubuntu required). Prerequisites: none on macOS or Windows; Linux needs curl and tar. Open a new terminal, confirm with `dragon --version`, then run `dragon` and paste your API key. Runs natively on macOS (Apple Silicon and Intel), Linux (x64 / arm64 glibc; Alpine/musl not yet), and Windows 10/11 x64 (ARM64 via emulation); Git Bash and WSL also work. **Do you upload or train on my code?** To make whole-codebase retrieval work, Dragon Code indexes your working repo and uploads the index to the gateway. That index is owner-scoped, secret files are filtered out before upload and again server-side, indexing can be disabled, the index can be deleted at any time, and MWS never trains on your code. **What are the Dragon lanes?** Dragon (default), Tiny (cheapest, `/tiny`), Large (`/large`), and Max (hardest tasks, `/max`). General work stays on Dragon. **What is Advisor?** An automatic escalation path: the fast model drives your session, and single hard questions go to a flagship frontier model. You pay frontier rates for that call only. **How much does it cost?** Plans from $5/mo (Noob, 10M tokens) to $750/mo (Supermax, 1B tokens). Every plan is a monthly Dragon token pool; top up with pay-as-you-go credits if you run past it. **Can I use Dragon from my own tools?** Yes. Point your Anthropic or OpenAI SDK at https://api.mws.run/v1 with an MWS API key. Usage bills pay-as-you-go from credits. ## URLs and structured data - Homepage: https://mws.run/ (sections: #dragon-code, #dragon, #advisor, #privacy, #pricing, #faq) - Install guide: https://mws.run/install - API overview: https://mws.run/api - API docs: https://mws.run/docs - System status: https://mws.run/status - Privacy policy: https://mws.run/privacy - Terms of service: https://mws.run/terms - Acceptable use policy: https://mws.run/aup - Architecture deep-dive: https://mws.run/architecture - Sitemap: https://mws.run/sitemap.xml - Short version of this file: https://mws.run/llms.txt The site serves JSON-LD structured data: Organization, WebSite, and SoftwareApplication (Dragon Code and Dragon) on every page; Product per plan and FAQPage on the homepage; HowTo on the install guide.