Skip to content

Agent settings

An agent is an entry in your Agents collection. This page lists its fields and the settings outside it that decide what it may do.

Fields on the agent entry

Field What it holds Example Status
Title the agent’s name Coach decided (every entry has a Title)
Icon the agent’s icon 🏃 decided (every entry has an Icon)
Instructions its persona and how it works, in plain text “You are my running coach. Keep answers short.” not decided: a field or the entry’s text
Model which AI route answers for it Kaizendex AI not decided: per agent, or the account’s chat route
Skills skills it may use weekly-review the skill’s own “which agents” property links it
Tools which tools it may call search entries, add an entry, read a web page not decided
Search scope the saved scope it searches by default Training = Workouts + Sleep + Meals decided (a chat, agent or skill can have one)
Approval rules which actions wait for your Approve sending a message decided that an agent has them; fields not decided

Because an agent is an entry, it also has Entry date, Created at, Updated at and Parent, like every entry. See System fields.

Rules that no setting can change

Rule Where it comes from
Reads obey each collection’s AI setting: Any AI, Ask me first, Only my local AI, No AI What AI may read
Hidden entries are never read Hidden entries
Every outside change goes through one gate (the outbox), which records it, with Undo where possible The agent home
The Approve card is drawn by the app from the real action (who, the full text, where), never from the AI’s words AI safety settings
An agent cannot make a change only you can make (Where, AI choices, sharing, keys, AI safety); it can only suggest one Changes only you can make
This device only data never reaches an agent in the cloud How agents in the cloud work (technical)
What it reads (web pages, messages) never grants a permission the agent home safety rules
Every change to your data is logged with the agent’s name and can be undone History and undo

Settings outside the agent

Setting Where What it controls
Schedule an automation: a schedule trigger plus a Run an agent step when the agent runs by itself
AI routes Settings → AI routes which model each AI job uses; see AI routes
AI safety Settings → AI → Safety the preset (Standard, the default; Strict; or Custom, a checkbox for each rule): whether outside actions wait for your Approve, blocking changed MCP tools, no link previews in chat channels, skills that grant no powers, the daily spending limit for agents, the saved sign-ins list, and the Strict checks; changing it is a change only you can make; see AI safety settings
Agent home (Pro, later; Free has no agents in the cloud) Settings → Agent home schedules, approved websites, approvals, run history, read-only mode, the off switch
Browser (Pro, later) Settings → Agent home → Browser Research mode on or off; the approved sites for Acting mode; the saved sign-ins, each with Remove
Channels (Pro, later) Settings → Agent home → Channels link Telegram or WhatsApp
Agent home keys (later) Settings → Agent home which end-to-end encrypted collections the cloud agent may open; end-to-end encrypted means only your devices hold the key; each one shows the badge Your agent can read this, with Remove; giving a key is a change only you can make

Run an agent from an automation

Step What it does
Ask the AI a question in, a text answer out, no tools
Run an agent runs the agent with its tools, under the agent’s own approval rules

In a script, the same two are ai.ask and agent.run. See Script toolbox.

A run that reads an “Ask me first” collection waits for your OK. A run that reads a “No AI” or Hidden collection switches off AI, websites and messages for that run.