Your Mac has a hidden LLM. What if you could use it to chat with your own thoughts?
Turns out macOS 27 has a built-in LLM that you can access directly through Terminal using fm chat.
Even more interesting, fm serve exposes it through an OpenAI-compatible API. No Ollama, no API keys, no separate model downloads (Refer here for LLM setup).
Imagine having a personal memory powered entirely by the LLM already sitting inside your Mac. No cloud, no subscriptions, and everything stays local.
The idea is simple: dump your random thoughts, notes, ideas, or anything worth remembering.
Gray Box (my own open source project) organizes them into a connected Markdown wiki that you can actually have a conversation with. It even traces answers back to your original notes.
Everything stays local in plain markdown.
Later, you can ask things like:
- “What was that idea I had last Tuesday?”
- “What did I decide about Project X?”
- “How is this related to something I wrote last month?”
Since Gray Box already uses LiteLLM, Apple’s built-in model should work with just a configuration change.
>!llm:!<
>!model_name: "openai/system"!<
>!api_base: "127.0.0.1"!<
>!api_key: "unused"!<
Curious if anyone else is experimenting with fm serve for something beyond regular chatting.
3 comments
Your Mac has a hidden LLM. What if you could use it to chat with your own thoughts?
Turns out macOS 27 has a built-in LLM that you can access directly through Terminal using fm chat.
Even more interesting, fm serve exposes it through an OpenAI-compatible API. No Ollama, no API keys, no separate model downloads (Refer here for LLM setup).
Imagine having a personal memory powered entirely by the LLM already sitting inside your Mac. No cloud, no subscriptions, and everything stays local.
The idea is simple: dump your random thoughts, notes, ideas, or anything worth remembering.
Gray Box (my own open source project) organizes them into a connected Markdown wiki that you can actually have a conversation with. It even traces answers back to your original notes.
Everything stays local in plain markdown.
Later, you can ask things like:
- “What was that idea I had last Tuesday?”
- “What did I decide about Project X?”
- “How is this related to something I wrote last month?”
Since Gray Box already uses LiteLLM, Apple’s built-in model should work with just a configuration change.
llm:
model_name: "openai/system"
api_base: "127.0.0.1"
api_key: "unused"
Curious if anyone else is experimenting with fm serve for something beyond regular chatting.
3 comments
@vijayshekhar The offline part caught my eye, though it’s Mac-only for now. I’m building “Gray Box” around a similar idea across platforms: dumped raw notes become a linked Markdown wiki you can question, with answers traced back to your notes. No embeddings needed.
Aaryanverma/graybox
4 likes · 1 repost · 1,085 views
Mentions
Your Mac has a hidden LLM. What if you could use it to chat with your own thoughts? Turns out macOS 27 has a built-in LLM that you can access directly through Terminal using fm chat. Even more interesting, fm serve exposes it through an OpenAI-compatible API. No Ollama, no API keys, no separate model downloads (Refer here for LLM setup). Imagine having a personal memory powered entirely by the LLM already sitting inside your Mac. No cloud, no subscriptions, and everything stays local. The idea is simple: dump your random thoughts, notes, ideas, or anything worth remembering. Gray Box (my own open source project) organizes them into a connected Markdown wiki that you can actually have a conversation with. It even traces answers back to your original notes. Everything stays local in plain markdown. Later, you can ask things like: - “What was that idea I had last Tuesday?” - “What did I decide about Project X?” - “How is this related to something I wrote last month?” Since Gray Box already uses LiteLLM, Apple’s built-in model should work with just a configuration change. >!llm:!< >!model_name: "openai/system"!< >!api_base: "127.0.0.1"!< >!api_key: "unused"!< Curious if anyone else is experimenting with fm serve for something beyond regular chatting.
Your Mac has a hidden LLM. What if you could use it to chat with your own thoughts? Turns out macOS 27 has a built-in LLM that you can access directly through Terminal using fm chat. Even more interesting, fm serve exposes it through an OpenAI-compatible API. No Ollama, no API keys, no separate model downloads (Refer here for LLM setup). Imagine having a personal memory powered entirely by the LLM already sitting inside your Mac. No cloud, no subscriptions, and everything stays local. The idea is simple: dump your random thoughts, notes, ideas, or anything worth remembering. Gray Box (my own open source project) organizes them into a connected Markdown wiki that you can actually have a conversation with. It even traces answers back to your original notes. Everything stays local in plain markdown. Later, you can ask things like: - “What was that idea I had last Tuesday?” - “What did I decide about Project X?” - “How is this related to something I wrote last month?” Since Gray Box already uses LiteLLM, Apple’s built-in model should work with just a configuration change. llm: model_name: "openai/system" api_base: "127.0.0.1" api_key: "unused" Curious if anyone else is experimenting with fm serve for something beyond regular chatting.
@vijayshekhar The offline part caught my eye, though it’s Mac-only for now. I’m building “Gray Box” around a similar idea across platforms: dumped raw notes become a linked Markdown wiki you can question, with answers traced back to your notes. No embeddings needed. Aaryanverma/graybox