Studio
Local LLM For Coding
Free forever with default model — no credit card required.
Product : avgidea.io/en/greenative
Read docs : gn-docs.avgidea.io
Run Claude Code or Codex on a local LLM
Greenative Studio now includes a local API gateway for agentic coding.
Point your coding agent at your Mac instead of the cloud. Your source code, your prompts, and the model itself stay on your machine.
-
Works with Claude Code / Codex and other compatible coding agents
-
One-click start — set host and port, pick a model, done
-
Run coding-focused models like NVIDIA's Nemotron 3.5 Lightning (32 GB+ Apple Silicon)
-
Live token metrics for every session
-
No per-token charges — your agent can code all day
Coding agent connected to Local LLM through Greenative Studio
Free plan available — no credit card required.
How it works
Install Studio, pick a model
Download Greenative Studio and a coding model such as Nemotron 3.5 Lightning or Qwen3 Coder. No Python, Homebrew or command-line setup.
Enable Coding Gateway
One click starts local Anthropic- and OpenAI-compatible endpoints on your Mac.
Point your coding agent at it
Set ANTHROPIC_BASE_URL (Claude Code) or OPENAI_BASE_URL (Codex) to the local endpoint. Your agent now talks to your local model instead of the cloud.
Why run your coding agent locally?
-
Real multi-file work
Refactors, new modules and test updates across a codebase — not just single-file snippets. -
No tokens, no API bills
One flat price. Run as many sessions as you like and watch live token metrics without a meter running. -
Works offline
On a plane, on a client site, behind a firewall. The model is already on your Mac. -
Your code stays on your Mac
No cloud inference. Source code, prompts and the model stay on your machine unless you connect an external tool yourself. -
Open-weight
Greenative Studio supports open-weight models with permissive licences only (Apache 2.0, MIT, OpenMDW).
System requirements
-
Mac : Apple Silicon (M1 or later), including cloud-hosted Macs such as Amazon EC2 Mac instances.
-
Memory : 16 GB or more for everyday use. 32 GB or more unified memory for coding agents and larger models such as Nemotron 3.5 Lightning — M2 Pro, M3 Pro, M4 Pro or better recommended.
-
macOS : macOS 14 Sonoma or later.
-
Disk : 20 GB or more free. 50 GB or more for coding agents and larger models.
-
Coding tools : Claude Code, Codex CLI, and other agents that speak the Anthropic or OpenAI API.
FAQ
Does it work with Claude Code and Codex?
Yes, both. Launch the Coding Gateway in Greenative Studio, then set ANTHROPIC_BASE_URL (Claude Code) or OPENAI_BASE_URL (Codex) to the local endpoint. Requests are handled by the local model regardless of the model name the client specifies.
How does it compare to Claude Code or Codex with the cloud API?
A local model is smaller than frontier cloud models, so it is best for routine edits, refactors and tests rather than large open-ended tasks. The trade-off is zero per-token cost and full privacy.
Does my code leave my Mac?
No. Inference runs on your machine. Your coding agent talks to a local endpoint; nothing is sent to Avgidea or any cloud unless you configure an external tool or MCP server yourself.
Which Mac do I need?
Any Apple Silicon Mac (M1 or later) running macOS 14 or later. For coding agents and models such as Nemotron 3.5 Lightning, 32 GB or more of unified memory and 50 GB of free disk are recommended. Cloud-hosted Apple Silicon Macs such as EC2 Mac are supported; stopping and starting an EC2 instance can change its hardware ID, and we offer a licence transfer through support.