Install
git clone https://github.com/hAcKlyc/cameoInstall command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →Cameo
An image-first canvas for your local Codex agent. Point at images, mark them up, and talk to Codex — generations and edits fan out across an infinite canvas.
cameo.ink · English · 中文 · License (AGPL-3.0)
What is Cameo?
Cameo is a native desktop canvas that gives your local Codex agent hands and eyes for image work.
Chat is a terrible interface for images. You can't point at the corner of a photo and say "this part." You can't see three variations side by side. You can't keep the original next to the edit. Codex can already generate and edit images — but a single scrolling chat column flattens all of that into a thread you lose track of.
Cameo puts that capability into a spatial canvas instead. You open a folder, its images lay out on an infinite board, and you work the way you actually think about images: select one, draw a box on the part you mean, type an instruction, and the result lands right next to the source so the lineage is visible. The conversation stays continuous — "warmer," "now change the background," "do that to these three" — because there's one ongoing Codex session per board.
Cameo is the hands and eyes, not the brain. It does not host a model, sell tokens, or orchestrate agents. All generation and understanding stay with your local Codex CLI, using whatever Codex credentials or provider configuration you already have. Cameo owns only what a chat UI structurally can't: spatial layout, file identity, non-destructive lineage, and "what you're pointing at."
Key ideas
- Board = folder. A board maps 1:1 to a local folder, which is also the agent's working directory. Your files stay yours, on disk, in a format any other tool can read.
- Non-destructive, always. Originals are never overwritten. Every output is a new image with visible lineage — position (left → right) is the history.
- Point, don't describe. Mark a region and it's sent to the agent as an image overlay — runtime-agnostic, no structured-mask API required. Any model that can see an image can use it.
- One continuous session per board. Codex holds the conversation (it's stateful and can ask clarifying questions); Cameo mirrors the timeline. Not per-image one-shots.
- Provide context, let the agent reason. Cameo injects what you selected and gets out of the way. No workflow state machines re-implementing intent.
Features
Similar plugs
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
Documented system prompts from Anthropic - Claude Fable 5.1, Opus 5.5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro,…
Wrap Antigravity, ChatGPT Codex, Claude Code, Grok Build, Muse Code, Devin as an OpenAI/Gemini/Claude/Codex compatible API service, allowing you to enjoy the free Gemini…