No foundation model included
You bring a provider, a subscription, an API key or a local model. Code Buddy is the agent and the routing layer.
Source-available · BSL 1.1 · Windows · macOS · Linux
A multi-LLM coding agent that runs with YOUR plans — or 100% local — at €0.
Code Buddy reads your repository, edits files, runs commands, tests the result and keeps going when the task is bigger than one prompt. Bring a subscription or API key, point it at Ollama, or connect a whole Fleet of peers.
15+ provider routes~110 tools27K+ testsFleet-ready
$ buddy try
one task · one agent · real tools
writes FizzBuzz + a test
runs the test
independently verifies it
cost$0.0000
Read the exact run →
15+provider routes
~110agent tools
27K+Vitest tests
3desktop platforms
Conservative core counts: 15 direct provider strategies; the extended catalogue lists 64 entries. See the scope →
The big picture
One runtime, any AI.
What sits between you and your projects: the interfaces, the agentic runtime, the tools and providers, and the loop that turns an idea into an improvement.
Real captures, no concept art
From a sentence to working software.
App Studio scaffolds a real project, starts a real dev server and shows a live preview. The same engine can write a test in the terminal or reason locally before it acts.
The output is real too
A landing page, not a mockup.
App Studio produced the editorial layout, copy, navigation and working preview from a plain-language brief. The puppy photos were generated locally; nothing in this capture was hand-coded for the showcase.
Every visual above is an existing repository capture. Open a GIF to watch it; open its link to inspect the source MP4 where available.
The first minute
Try it in 60 seconds.
Use your existing ChatGPT Plus/Pro login, or stay on your machine with Ollama. The command at the end runs the real FizzBuzz proof.
Read the getting started guide →01
npm install -g @phuetz/code-buddy02
buddy login # or: ollama pull qwen3.6:35b-a3b03
export CODEBUDDY_PROVIDER=ollama # omit after login04
buddy tryOne engine, many surfaces
What it actually does.
Terminal, desktop, server and peers share the same agentic loop. Click into the source docs for the gates, commands and boundaries behind each card.
01 / build loop
Coder · edit · test
Reads a repo, edits files, runs commands and verifies the result through a stateful multi-turn loop.
02 / desktopApp Studio
Describe an app; get real npm install, a dev server, live preview, capped auto-fix and GitHub push.
03 / routingMulti-LLM + council
Switch providers, fail over, ensemble answers or ask several models to critique and synthesize.
04 / controlLocal · private
Run Ollama or LM Studio with permission modes, sandbox tiers, secret handling and fail-closed guards.
05 / networkMulti-agent Fleet
Peers stream events and call each other’s models and read-only tools across your own network.
06 / extensionMCP client + server
Connect external tools and expose Code Buddy to other hosts — with the same permission boundary.
07 / mediaVideo / film studio
Prompt → scene plan → clips → Piper narration → karaoke captions → an ffmpeg montage.
08 / cockpitCowork desktop
An Electron workspace for chat, files, workflows, fleet, reasoning, App Studio and Video Studio.
No hype required
The honest version.
Code Buddy is broad because it is an open engine, not a hosted model. Here are the trade-offs before you install it.
Local depends on your hardware
Ollama at €0 is private and unmetered, but speed and quality depend on the model and machine you choose.
More surface means more setup
The Fleet, desktop and integrations reward configuration; the community is smaller than the established alternatives.
Short comparison
Full comparison →| Capability | Code Buddy | Claude Code | Codex CLI | Aider |
|---|---|---|---|---|
| Multiple providers | ✓ 15 routes / 64-entry catalogue | — Claude models | ◐ OpenAI + compatible/local | ✓ via LiteLLM |
| Local / offline | ✓ Ollama, LM Studio | — network model | ✓ OSS + local | ✓ local endpoints |
| MCP | ✓ client + server | ✓ client + server | ✓ client + server | — no native docs |
| Peer Fleet | ✓ multi-machine mesh | — not multi-provider | — not peer mesh | — |
| Desktop | ✓ Cowork | ✓ desktop | ✓ separate app | ◐ experimental web UI |
A “—” describes a documented capability, not product quality. The source matrix was checked on 2026-08-23; offers and quotas change.
Open source, inspectable, yours
Give your next task a real agent.
Start with one command. Keep the model on your machine, use the plan you already pay for, or put several agents to work together.
Built in public