Terminal & Cowork
The identical execution engine is embedded in the CLI terminal and the Cowork desktop app.
AI Studio · Open source · MIT License
Code Buddy: the coding agent :
on your local machine.
An autonomous agent that reads, edits, and executes code directly on your workstation, using the LLM provider you already pay for: ChatGPT Plus, Claude Pro, or fully local open models. No third-party API tokens to purchase, no per-token markup.
Capabilities
A coding agent, a desktop studio, and your existing models.
Code Buddy is engineered for software developers wanting to work with their preferred provider, product teams prototyping reactive interfaces locally, and security-conscious engineering departments that require their codebase to remain on premise.
Fully Tooled Agent
Code Buddy reads, writes, and executes your code with 200 registered tools and persistent memory.
Your Existing Providers
Fifteen providers, including free tiers and local offline models (Ollama), run with your existing accounts.
Local Research & Production
Cited technical document search, local ComfyUI video rendering, and multi-instance fleet management.
A single engine, any foundation model.
What sits between you and your codebases: multi-channel client interfaces, agentic coordination engine, runtime tools and providers, and the autonomous loop translating concepts into tested pull requests. Infographics dated September 10, 2026, metrics from published README.
Terminal and Cowork, captured in their actual state.
The screenshots below were recorded from live runs on September 2, 2026. 'Atelier Nord' datasets are fictional. Empty interface states remain shown as they are.
Terminal · 01
Generate and test directly from the shell
A live session creating inventaire.json and inventaire.mjs, adding the --total flag, and running the script against sample furniture data.
Terminal · 02
Inspect tool permissions and security profiles
The buddy tools profile command details the fleet.hermes.balanced profile, inspected capabilities, and active sandboxed allow-lists.
Terminal · 03
Verify runtime prerequisites with buddy doctor
buddy doctor --verbose confirms runtime readiness to interact with local Ollama instances and outputs an exhaustive setup diagnostic.
Terminal · published demonstration
Interactive trial archive.
The initial demonstration GIF remains available: it shows buddy try generating and testing a FizzBuzz algorithm.
Cowork · 01
A unified developer desktop environment
Cowork integrates Chat, Planning, Activity monitoring, File management, App Studio, and Video Studio into a native desktop suite.
Cowork · 02
Interactive dialogue with the agent
This live session connects to gpt-5.6-sol using the Atelier Nord test data. It highlights the dual chat feed and activity inspector.
Cowork · 03
Scaffold interactive web applications
App Studio exposes generative prompts, live React/HTML runtime previews, and a curated blueprint library. This view captures the honest initial state.
Cowork · 04
Orchestrate generative video production
Video Studio displays the timeline canvas, four empty scene slots, 16:9 framing, a 6-second duration, and ComfyUI integration. No synthetic video was rendered for this capture.
Established Cowork demonstrations remain published.
Getting Started
Install Code Buddy.
The terminal CLI and the desktop Cowork suite share the identical core agent. Configure the LLM provider and execution environment tailored to your requirements.
Terminal · npmopen source
# Install globally
npm install -g @phuetz/code-buddy
# Run health diagnostics or start interactive session
buddy doctor
buddy36,154
tests · working branch · 09/03
103
verified CLI commands
3,018
Cowork automated tests · 09/02
Honest Limitations
What Code Buddy does not claim to be.
The tool is documented with transparency regarding its production boundaries.
Engineered for software developers
Requires command-line proficiency and an active LLM provider (commercial account or local inference server).
Experimental advanced architectures
Autonomous self-improvement cycles and multi-model consensus councils are active research experiments labeled as such.
Code Explorer Advisory
Audit and untangle legacy software assets.
Codebase topology mapping, blast-radius assessment, architectural bottlenecks, and an actionable remediation roadmap. First commercial engagement delivered in 2026. Let's discuss your legacy challenges.