any model, one terminal
drive anthropic, openai, gemini, azure, bedrock, vertex, or a fully local model. switch providers with one command — no code changes, no lock-in.
open source · model-neutral · verified no phone-home
runs anywhere.
uses anything.
not a chatbot wrapper or another ide plugin. an open coding agent that runs in your terminal, talks to any model, and keeps every change reviewable.
node ≥ 22 · macos, linux, windows, android · works offline with local models
░█▀▄░█▀▄░█▀▀░█░█
░█▀▄░█▀▄░█▀▀░█▄█
░▀▀░░▀░▀░▀▀▀░▀░▀
░░░█▀▀░█▀█░█▀▄░█▀▀
░░░█░░░█░█░█░█░█▀▀
░░░▀▀▀░▀▀▀░▀▀░░▀▀▀
open terminal for any LLM · vx.x.x
› ▌
one terminal. every provider.
- hicap
- codex
- anthropic
- openai
- gemini
- aws bedrock
- google vertex
- azure
- github copilot
- ollama
- lm studio
- openrouter
- together ai
- litellm
- deepseek
- kimi
- qwen
- grok
- +200 more
# features
a real agent loop with real tools — not autocomplete with extra steps.
real tools, not just chat
reads and edits files, runs shell commands, searches your repo with ripgrep, browses the web, and spawns subagents — all streamed live to your terminal.
sandboxed by default
bash commands run inside a sandbox with a per-tool permission system. plan mode, git worktrees, and streaming diffs keep every change reviewable.
profiles per repo
a .brewcode-profile.json pins the model, base url, and runtime defaults, so every clone of your repo boots the exact same agent.
streaming, not batch
watch thinking, tool calls, and diffs render token-by-token in a fast react-powered tui — with themes, vim keybindings, and a statusline.
extensible everything
mcp servers, skills, hooks, 150+ slash commands, scheduled tasks, swarms, and workflows. plus a grpc server and editor modes for your own tooling.
# verified no phone-home
every release build is mechanically scanned before it ships. if a single telemetry endpoint, metrics call, or infrastructure fingerprint shows up in the bundle, the build fails.
- no telemetry or usage metrics endpoints
- no environment or infra fingerprinting
- no feedback back-channels
- local models via ollama or lm studio never leave your machine
your code and prompts go only to the model provider you choose. nothing calls home.
$ bun run verify:privacy
scanning dist/cli.mjs …
✓ no telemetry endpoints
✓ no metrics batching
✓ no infra fingerprinting
✓ no feedback channels
✓ no banned patterns in build output
# bring your own model
cloud, gateway, or fully local — brewcode speaks them all natively.
cloud
hicap, openai & codex (chatgpt oauth), anthropic (direct, bedrock, vertex, foundry), azure openai, google gemini, github copilot, deepseek, kimi, qwen, grok, glm, and more.
gateways
route through openrouter, together ai, or litellm for policy, cost control, and failover — pick a goal like latency, balanced, or coding and let the profile choose.
local
ollama and lm studio work with zero api keys and zero network. ollama launch brewcode and you're coding offline with llama, qwen, or deepseek-coder.
# same agent, native app
brewcode desktop is a native tauri 2 shell around the exact same agent loop as the cli — one brain, two frontends.
- chat, reviewer, and require-plan modes
- live streaming of thinking, tool activity, and diffs
- model & provider picker, workspace switcher, git branches
- plan / files / terminal inspector tabs
- macos, windows, and linux
# install
three commands from zero to agent.
-
install the cli globally
-
open any repository
-
start brewing
-
point brewcode at your local ollama server — no api key needed
-
or launch straight from ollama
-
your code never leaves your machine
-
clone and install (bun 1.3+)
-
build with the privacy gate
-
link it globally
-
live on the edge with the nightly dist-tag
-
go back to stable any time