Analysis updated 2026-05-18
Chat with an AI model from the terminal using your existing Claude or OpenAI login
Delegate a task to multiple sub-agents that work in parallel and report back
Run untrusted or AI-generated code safely inside a sandboxed environment
Watch multiple active agents visualized as avatars in a 3D world interface
| ajensenwaud/varela | 3imed-jaberi/cryptography-si-isamm | 3imed-jaberi/koa-isomorphic-router | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | 2021-09-25 | 2021-02-06 |
| Maintenance | — | Dormant | Dormant |
| Setup difficulty | moderate | easy | easy |
| Complexity | 4/5 | 1/5 | 2/5 |
| Audience | developer | researcher | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 20+ and either an OAuth login or API key for Anthropic, OpenAI, or OpenRouter.
varela is a command line AI agent tool built in plain JavaScript with no external dependencies at all. It lets you chat with an AI model, connects to Anthropic, OpenAI or OpenRouter depending on which login or API key you set up, and can carry out tasks by calling tools on your behalf. Its main distinguishing feature is that it can rewrite parts of its own source code while it is running and reload the changes immediately, and it logs what worked and what failed so it can turn wins into new tools and failures into lessons over time. Running it only needs Node.js version 20 or newer. There is no separate install step in the usual sense: the folder you download is the running program, and using the built in self-editing feature changes files inside that same folder. Small installer scripts are provided just to add a shortcut command to your system so you can type varela from anywhere. The tool can run in several modes: a full-screen terminal chat, a split-view interface showing a file tree alongside the chat, a dashboard showing agent activity, a one-shot mode for a single task, and a background daemon mode that other clients can connect to over a local WebSocket connection. There is also an optional 3D visual interface where each active agent appears as an avatar in a virtual world, useful for watching multiple agents work at once. For safety, code the agent generates or otherwise does not fully trust runs inside a sandboxed environment with no file or network access by default, separate from the tool used for normal trusted automation. The chat interface supports features like pasting images for the AI to look at, attaching files by typing an at sign followed by a filename, running shell commands directly without going through the AI, and different permission levels ranging from read only planning to fully automatic file edits. The project can also delegate a task to multiple sub-agents that work on separate pieces in parallel and report back once finished.
A zero-dependency, self-modifying command line AI agent that chats with Claude, OpenAI or OpenRouter models and can run in a 3D visual world.
Mainly JavaScript. The stack also includes JavaScript, Node.js, WebSocket.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.