Settings reference
Open Settings from the navigation. It’s organised under four tabs —
General, Agents, Supporter, App — and everything is also
written to a plain config.toml you can edit by hand.
General
Wikis
- Wikis root — the folder that holds all your wikis. Changing it requires restarting Wikai.
- Default wiki — optionally jump straight to one wiki on launch instead of the picker. Manage wikis and + Create new live here too.
Interface
- Open the chat sidebar by default — whether the sidebar opens automatically.
Lint thresholds
- Stale-after days — how old the gap between edit and
updatedmay be before a page is flagged (default 30). - Minimum cross-reference length — shortest title/alias match to flag (default 4 characters).
Agents
One sub-tab per agent, each with its brand mark and a status dot: Claude Code, Codex, Anthropic API, OpenRouter, llama.cpp, vLLM, and a Grok tab marked coming soon. The three least-exercised carry a small beta tag. What each tab holds depends on what that agent needs — see The chat sidebar for what each one is.
- Claude Code and Codex — the detected binary (path, version) with a
Re-detect button and a Use a different binary override, and a
reminder to sign in once in a terminal (
claude /login,codex login). - Anthropic API and OpenRouter — an API key, and a model. Note: keys
are stored in plain text in your
config.toml(a move to the OS keychain is planned). - llama.cpp and vLLM — the server’s address (and a key, if you started it with one), the model, a line saying where your pages go given that address, and a Tool set.
The model can also be set from the dropdown beside the chat — it’s the same setting.
Tool set — how many of Wikai’s MCP tools a local model is
given. auto (the default) chooses by the model’s context window: reads
under 16k, core from 16k to 32k, all from 32k up — and reads when the
server doesn’t report a window, which plain llama-server doesn’t; an Ollama
server is asked. Set reads (the five read-only tools), core (those
plus create page, edit block, and work documents — nine) or all by hand
for a server that reports nothing, or to try a smaller set on a model that
struggles with many tools.
Default chat backend — which agent a new chat starts with, or Last used: whichever agent you last picked in the sidebar, remembered across restarts and across wikis. The line beneath says which one that currently is.
Daily note enrichment
How the agent handles your captures — classifying them, suggesting links, tidying shorthand:
- Automatic — new captures are read as they land.
- Manual — only when you click Enrich now in the Today view.
- Off.
Automatic is unavailable on a metered agent — the Anthropic API and OpenRouter — where it would spend your credits on a timer you never set; use Manual there. A local model (llama.cpp, vLLM) is free to run, so Automatic is allowed. Either way, enrichment only runs while you have the day’s note open, and nothing is written to your wiki without a click.
MCP server
The local endpoint external agents connect to. They need its bearer token as well as the URL; Wikai’s own chat agents handle this automatically.
- Copy copies the endpoint URL only; it is not enough to authenticate.
- Show / Hide reveals or masks the MCP bearer token; Copy token copies it.
- Generate new replaces the token on the spot, after a confirmation — no restart. External clients configured with the old token stop working until you paste the new one; Wikai’s own agents switch by themselves.
- Copy client config (JSON) copies the endpoint and authorization header
together in an
mcpServersconfiguration — the one thing to paste into an external client. The snippet contains a secret, so keep it out of shared screenshots and support reports.
The token is saved in config.toml and kept across restarts until you
generate a new one.
- Preferred port — chosen on first launch (27182, or the next free port) and saved, so an external client’s configuration survives restarts. Type a port to choose a specific one; Use automatic clears the field, and the next launch chooses again from 27182 up and saves the result. If the saved port is taken at launch, Wikai moves to the next free one, saves that, and says so — a notice here and a toast at launch, naming both ports.
See MCP.
Supporter
Your licence state: whether Supporter is active, which machines you’re signed in on, and a way to sign out of one. Activate by email link or by pasting a licence key. See pricing.
App
Updates
The version you’re running, a Check for updates button, and Check for updates on launch — on by default, and a real off switch. With it off, Wikai doesn’t contact the update server until you ask it to. When a newer version is found, an Install now banner appears at the top of the window.
Keyboard shortcuts
Opens the same cheat sheet as ⌘/. See keyboard shortcuts.
Diagnostics
Where Wikai writes its logs, with a button to open the folder. Worth attaching to a bug report.
Two lines about memory sit below it. Memory is Wikai’s own footprint — the number Activity Monitor or Task Manager shows — with its peak, whether the embedding model is loaded (it unloads after ten idle minutes), and what the app is doing; Refresh re-reads it. System memory is the machine’s side: how much is available, the operating system’s pressure level, and on macOS how much of Wikai has been compressed out. When an ingest is slow on a starved machine, these two lines tell you whose fault it is.
About
The version, and the open-source software Wikai is built from with each project’s licence.
Where config lives
macOS ~/Library/Application Support/io.atomicbits.wikai/config.toml
Linux ~/.config/io.atomicbits.wikai/config.toml
Windows %APPDATA%\io.atomicbits.wikai\config.toml Last updated 2026-10-07.