Guace

macOS · local-first · your keys, your files

Your AI-native knowledge base — that you can talk to.

Every note, conversation, and decision lives in your own markdown vault on your own disk. Chat with it in plain language — and when a job needs a real worker, Guace dispatches it to managed agents like Claude Code and Codex, then folds the result back into your vault.

Guace dashboard — your vault, a checklist, and an agent chat panel beside it
Markdown vault — file tree on the left, a live-rendered note in the center
A managed-agent task that ran in its own isolated workspace
Usage dashboard — token, cost, and prompt-cache breakdown from real provider data
An HTML file rendered inline inside the vault, next to the file tree

What it does

A vault you can chat with — and agents that do the heavy lifting.

The "AI-native" in AI-native knowledge base isn't only the agents. It's that your notes are something you talk to, not a folder you have to grep.

01

AI-native, end to end

The built-in agent isn't bolted on. From one conversation it acts directly across everything in Guace — your vault, managed agents, todos, and tasks.

02

Local-first

Your notes are plain files on your own disk and your keys live in macOS Keychain. Requests go straight to your provider — no SaaS in the middle, no telemetry.

03

Powerful managed agents

Hand heavy jobs to real local workers — Claude Code and Codex today, more runtimes coming. Each runs in its own isolated workspace and reports the result back.

04

Transparent token & cache data

Know exactly what you spend. Accurate per-model token counts and prompt-cache hit rates, straight from real provider usage — never estimates.

05

Bring your own provider

Point Guace at the LLM backend you want and plug in your own keys. Swap or add providers whenever you like — no lock-in.

06

Previews every file

Markdown and HTML render inline, plus PDFs, images, and plain text — .md .html .pdf .txt and more, right next to your notes.

Why you might want it

Not a SaaS. Nothing about your knowledge leaves your machine.

  • You already keep markdown notes and want an agent that can actually edit them — Obsidian-compatible, plain .md files.
  • You use Claude Code / Codex daily and want a chat-style manager that decides when to delegate to them.
  • You want the data, the prompts, and the keys to stay on your laptop — keys in Keychain, requests straight to your provider.
  • You've felt the pain of context windows and want a tool that's honest about cache hits, token cost, and what it's actually doing.