Compatibility

This matrix separates environments exercised by automated tests from those that are expected to work but still need community verification.

Application platforms

Platform	Status	Notes
Ubuntu Linux, Node.js 20 and 22	CI tested	PGlite and PostgreSQL migrations, web build, unit tests, and Playwright run in GitHub Actions.
macOS on Apple Silicon	Maintainer tested	Primary development environment; embedded PGlite requires no separate database service.
macOS on Intel	Expected	Supported by Node.js and PGlite; community verification is requested.
Windows 11	Experimental	Run from PowerShell or WSL2; clean-install verification is still needed.
Other Linux distributions	Expected	Requires Node.js 20.9+.

The PoseForge application itself does not require a GPU. The selected image engine determines compute and memory requirements.

Engine	Required setup	Execution location	Local GPU	Notes
Codex CLI	Installed and authenticated `codex` CLI	Model provider	No	Uses the signed-in Codex account; selected references and prompt leave the machine.
Google Antigravity CLI	Installed and signed-in `agy` CLI	Model provider	No	Uses the signed-in Google plan; selected references and prompt leave the machine.
ComfyUI	Running loopback server, models, and API workflow	Your machine	Usually	Hardware needs depend on model, resolution, precision, and workflow.
OpenAI	API key with image-model access	OpenAI	No	API usage may be billed separately from a consumer subscription.
Google Gemini API	Gemini API key	Google	No	API access and billing are separate from Antigravity CLI authentication.
Replicate	API token and model access	Replicate	No	Model capabilities and pricing depend on the configured model.
fal.ai	API key	fal.ai	No	Uses the configured hosted image-editing model.

These are starting points, not guarantees; custom nodes and models can change requirements substantially.

Hardware	Suggested starting point
8–12 GB VRAM	SDXL/OpenPose workflows at moderate resolution; use memory-saving options where needed.
16 GB+ unified memory on Apple Silicon	Quantized/offloaded workflows may work, usually slower than a dedicated CUDA GPU.
16 GB+ VRAM	Larger image-editing workflows and higher resolutions become more practical.
CPU only	The application runs, but local generation may be impractically slow. Prefer a signed-in CLI or cloud engine.

For a compatibility report, include the information requested in SUPPORT.md. Do not include private photos or credentials.