Cmd+, (Ctrl+, on Windows and Linux) or run /settings in the chat. The sidebar groups the sections into four clusters: the app itself, what Kepler uses and may touch, what it knows, and labs. This page walks the sections in the order they appear in the sidebar. Where a topic has its own page, this page stays short and links to it.
App
General
Appearance
Plan & Usage
Kepler runs locally and is free: no seats, no cloud, no limits. This section shows your plan and what this month has cost so far in model spend, updated as you work. Cloud plans add team sync, shared memory and hosted automations. Below it, Activity lists recent work on three tabs: Runs, Tool calls, and Hooks (every hook run over the last 30 days). A run you cancel still counts the tokens it spent.Models and access
Models
Kepler runs on two models: a capable main model for reasoning and chat, and a fast worker model for the background lanes it spins up. Pick either from the composer’s model picker or with/model. Whichever model you pick is the one that runs, every turn. Kepler never switches models on its own: if a call fails, it retries the same model or tells you.
Reasoning effort is a slider in the same picker, or /effort. Auto lets the model decide.
Two optional picks sit beside them. The image model draws images when your chat model cannot. The decision model answers typed questions in about a second, and Kepler uses it to double-check commands before they run. It is off by default. See Decision model.
Providers. Enable a provider, add its key, and its models appear in the picker. Keys stay on this machine.
The picker shows how each model is billed. Background work, such as watchers and memory, never runs on a subscription plan like ChatGPT’s.
Context window.
You can also compact by hand at any time with
/compact.
Tools & MCP
Customize
One place for plugins, skills, hooks, slash commands and specialist agents, with one search, a chip per kind, and a workspace filter. The Agents chip also sets the model for each of Kepler’s own agents: Kepler itself, the automation builder, Memory, Topology and Shadow. See Customize.Permissions
What you have allowed for good, your command rules, the always-blocked list, and the Files, Shell and MCP servers switches, each set to allow, ask or block. See Permissions.Secrets
Named credentials for MCP servers, stored in your operating system keychain and referenced as${NAME}. Each row shows which integrations use it. See Extend Kepler.
Gateway
Connect WhatsApp, Slack, Telegram, Discord or a channel from a plugin so you can carry on a session from your phone. Per channel: what it can do (Observe, Ask before changes, Act within bounds), default workspace, reply in channels on @mention, show typing, read receipts, and who can message. Add channel installs a channel plugin, wraps a channel server you already have, or has Kepler build one. Disconnect everything turns off every channel and keeps the people you allowed. Reachable only while Kepler is running. See Chat from your phone.Remote access
Switch Work on other machines on to let a thread work on a server, a cloud VM or a sandbox. Below it: the machines Kepler knows, each with Connect, Disconnect or Start and a per-machine set of limits; commands running in the background; and New threads start on. See Remote access.What Kepler knows
Memory
Import, export, reindex, rebuild the graph, wipe, and read shadow briefs. See Memory.Watchers
Every watcher across every session, with counts for watching, checks, alerts sent and ended, and the detail of each one. See Watchers.Topology
See Topology.
Labs
Experiments
Settings that talk to the agent show “Can’t reach Kepler” if the app is still starting. Wait a moment and reopen the page.
Where to go next
Permissions
The rules layer under the posture.
Memory
What Kepler remembers and how to edit it.
Chat from your phone
Gateway setup for each channel.
Troubleshooting
When a setting does not take.