A free and fully featured GenAI frontend for PowerPC Macs.
Sage is a fully featured generative AI frontend designed specifically for PowerPC Mac. It provides a native, lightweight interface for interacting with AI models on legacy hardware. It does not run AI models locally . Instead, all inference is handled remotely through API endpoints, with responses streamed token-by-token to the client in real time.
Sage allows you to connect to most OpenAI-style APIs (including providers like Anthropic and Google) and supports custom endpoint selection for local inference servers like LM Studio, meaning you can bring-your-own-model (BYOM) as a cost-free alternative to using API credits.
Unlike other frontends for PowerPC Mac, Sage includes a custom text renderer that fully parses and displays model output as Markdown, supporting (most!) Markdown features. It also provides syntax highlighting for 12 common languages, inline image support, and the ability to upload any file as input.
We currently use the Responses API convention for most conversations and plan to support more advanced features like web search, stateful conversations, and better tools.
Sage supports Mac OS X "Leopard" 10.5.8 and, experimentally, the PowerPC Snow Leopard alpha. Mac OS X "Tiger" 10.4.11 and lower are not supported.
On Mac OS X Leopard, the built-in networking tools are too outdated to communicate with modern HTTPS endpoints. This makes it impossible to connect to current AI backends using the system utilities alone.
To resolve this (and prevent more hacky, less secure workarounds), Sage requires a backported, modern version of curl with up-to-date TLS and certificate support. This is provided via MacPorts/PowerPC Ports, which maintains newer software and development tools for PowerPC-based Macs.
Before continuing, ensure curl is installed through PowerPC Ports by using the bootstrap package provided here (Leopard) or here (Snow Leopard alpha). See the PowerPC Ports website for more info.
Note: At the moment, curl must be present in /opt/bootstrap or /opt/local.
Eventually, this version of curl will be bundled with Sage so that this is no longer a required step.
Once curl is installed, download the app, extract the ZIP, move "Sage" into your Applications folder and run. A Quick Start guide will open on your first launch explaining how to add an API key.
Add an API key from a provider of your choice, ensure you have the correct provider selected, and send a chat! You can invoke image generation requests by just asking the model, no special command needed.
Sage is designed as a chat-first assistant with optional agentic features available for power users or developers.
The primary goals are to:
- Extend the useful life of PowerPC Macs by providing a fast, native interface to modern AI and web services - no browser required.
- Lower the barrier to entry for PowerPC development and experimentation by giving developers immediate access to research and troubleshooting assistance tailored to this obscure and under-documented platform.
- Accelerate developer workflows by assisting with code review, crash log analysis, porting guidance, and diagnostics, with an optional ability to automate actions directly on the system.
- Enable trustworthy information retrieval for tasks such as technical research, news summaries, reference lookups, and/or factual lookups, putting a priority on verifiable citations instead of vague model assertions.
The first release, 0.2a is provided as an alpha for testing purposes and to gather feedback. I'm planning on releasing the source code at some point in the near future.
A list of planned features includes:
- Full MCP support for tool integrations
- Stateful server-side conversations
- Web search tool integration (defer to built-in tool for OpenAI models, create a custom DuckDuckGo search layer for custom/other models)
- Video generation
- Agentic code editor with support for diff editing, reading/writing files directly on system
- More syntax highlighting support
- No Tiger support.
- Inferencing is still being done off-device, Sage is only a frontend.
- BYOM approach requires paid API credits or your own GPU to power requests.
- G3s are not officially supported and will probably be super slow.