Hi, I built Loom Code, an open-source (MIT) AI coding agent that runs in your terminal. Repo: https://github.com/toshalkumbhar... Install: npm install -g loom-agent Why another agent? I wanted one that: - isn't locked to one vendor — same agent loop across Anthropic, OpenAI, NVIDIA NIM, Gemini, OpenRouter, or a local Ollama model (180+ models), switchable mid-session with /models - has a real terminal UI (custom OpenTUI + SolidJS renderer): chat with streaming, inline diffs per tool call, plan/build/chat modes, mouse selection + copy, vim mode, a command palette (ctrl+p), and a sidebar - ships the pieces I missed in one package: MCP servers (built-in browser + presets for Playwright, Context7, etc.), subagents you can invoke with @agent, LOOM.md project memory, usage/cost tracking per session and lifetime, session save/fork/undo, and a browser interface (`loom web`) Engineering notes for the curious: the TUI is SolidJS compiled to a custom terminal renderer (no HTML, no node canvas); bun is bundled via npm optional deps so `npm i -g loom-agent` needs zero setup and still works if you never install bun (falls back to a Node REPL); and there's an Agent Client Protocol server (`loom acp`) so editors like Zed can drive it. The bash/edit/write tool calls go through a permission system (allow/ask/deny per tool + path globs), and dangerous commands always ask. Breaking things is expected — it's early and rough around the edges. What would you want in a terminal coding agent? Contribute to toshalkumbhar8979-design/loomcode development by creating an account on GitHub.

loomcode

Loom Code – open-source AI coding agent for the terminal

Loom Code – open-source AI coding agent for the terminal