Your preferences
Settings → Preferences holds what is yours alone. Nothing on this screen affects another member, and nothing here is borrowed from a shared chat or a Project.
Appearance
Choose light, dark, or follow your system. Kaizen's themes are token-driven, so every surface — including these pages, the chat canvas, and diagrams — switches together.
Every combination is held to the same legibility bar: body text, muted text, links, status colors, and the label on a filled button all meet the WCAG AA contrast minimum in each palette, in both light and dark. Picking the palette you like will not cost you readable text.
If your system asks for reduced motion, Kaizen honours it everywhere — animations resolve immediately instead of playing.
Kamiwaza — the corporate palette, teal on black and white — is the default accent palette until you choose another. Your deployment may ship an organization theme instead; it becomes the starting palette for that deployment. A palette you choose is saved in this browser and wins over either default. Your light/dark choice applies on top of every palette.
Kaizen 0.4 intentionally starts a new browser-storage namespace. Preferences saved by the pre-0.4 browser bundle—including palette, recent searches, pins, and per-chat model choices—are not imported. Choose them once in Kaizen 0.4 and they persist normally from then on. The sole safety exception is an explicit artifact auto-run opt-out: Kaizen carries that disabled choice forward once so the release transition cannot silently re-enable automatic code execution.
Your circle
Every member is shown as a small circle beside their name — in the people strip at the top of a shared chat and in the roster behind it. By default the circle carries the first letter of your name.
Under Your circle you can upload a picture instead. Choose Upload a picture, or drop a picture onto the control. Kaizen then shows your picture behind a round mask, which is exactly the part that will be kept: drag the picture to move it, use the zoom slider to close in, and watch the two previews underneath, which are the real sizes a chat row and this page draw. When the framing is right, choose Use this picture. If you would rather not drag, the round area takes the arrow keys.
A picture has to be a PNG, JPEG or WebP file no larger than 4MB. A square picture of at least 512 pixels on a side gives the best result. Kaizen refuses anything else and tells you why, and your circle stays exactly as it was.
Your picture is stored as a square: Kaizen re-encodes it and keeps two sizes, one for the settings page and a small one for chat rows. Camera information traveling inside the file, including any location a phone recorded, is discarded and never reaches anyone else's screen.
Remove picture takes the picture away and puts your letter back. Under Letter color you choose the color that letter sits on; it is what people see whenever you have no picture, so it is worth setting even if you have one.
Every control here is an ordinary button, so you can reach it with Tab and choose with Enter or Space.
Your circle makes your own contributions easy to pick out in a busy shared chat. It is decoration only: your name is always written beside the circle, so a picture never becomes the only way to tell who someone is. The choice is yours alone — you cannot set anyone else's circle, and nobody can set yours.
Kaizen used to offer eight symbols here. It no longer does. If you chose a symbol while they were offered, it still shows everywhere and stays until you change it, and Your circle gives you a one-click way to move to your letter whenever you want.
Model and effort
Set the default model and effort level for the turns you start.
Effort is a reasoning budget: higher effort buys more thinking at the cost of time and tokens. Both defaults can be overridden per conversation from the controls beside the composer, so the preference is a starting point rather than a lock.
Effort does not promise a longer visible Thinking transcript. Chat can display only the reasoning narrative and streaming cadence that the selected model provider returns; private model reasoning is never exposed.
Which models are available is a deployment decision. See Models & embeddings if the list is not what you expect.
Tool approvals
Choose your default for when agents may use write-capable tools. Use workspace default follows the administrator's starting mode; choosing another mode changes only your turns. Set an individual agent's default from that agent's detail page, and override the current conversation beside the Chat composer.
| Mode | Behavior |
|---|---|
| Ask when needed | Kaizen pauses only for tools that declare that they need confirmation |
| Ask before writes | Kaizen pauses before every write-capable tool call |
| Read only | Write and sensitive tool calls are blocked |
| Never ask | The built-in workspace default. Permitted writes run without a Kaizen permission prompt |
An administrator can change the workspace default. Your saved personal, agent, conversation, or task choice still wins at its corresponding scope.
Never ask requires an explicit warning confirmation. It bypasses only the member-controlled permission prompt: agent tool grants, capability ceilings, resource permissions, sandboxing, administrator policy, audit, and MFA remain in force. Every prompt uses the same approval review surface; a tool may add member-safe fields, but it cannot substitute a separate confirmation channel or treat a general permission as consent to changed arguments.
A task can inherit the workspace, personal, and per-agent choice or store its own stricter or more autonomous mode. See Tasks & scheduling.
Chat compression
Three opt-in efficiency tools plus a master switch. Every tool defaults to off and stays off until you turn it on.
| Control | What it does |
|---|---|
| Member compression | Master switch. Turning it off pauses every tool without erasing their saved positions |
| Tool-result compression | Shortens eligible sandbox command output before the model sees it |
| Plain-speak responses | Asks Kaizen to write visible thinking and final answers more tersely |
| Prompt trim | Removes filler from your message before the model sees it, and keeps your chat showing what you wrote |
Turning any compression tool on while Member compression is paused resumes the master switch, so the selected tool applies to your next turn.
The same settings appear beside any Chat composer under Compression. Both surfaces save the same member preference, and a change applies to your next turn. See Chat compression for the detail, including what is deliberately never compressed.
What is not here
Some things people look for on a preferences screen live elsewhere on purpose:
- Notification categories — managed from the notification inbox.
- Sub-agent selection — per agent, not global, so it lives with the agent. See Sub-agents & delegation.
- Per-agent tool approvals — open the Chat agent in Agents and choose its default beside its capabilities.
- Deployment configuration — models available, connectors, tools, and feature availability are administrator settings, not member preferences.
Privacy of your settings
Your preferences are member-owned. A shared chat never adopts your settings for someone else's turn, and never applies theirs to yours — each turn uses the state of its own initiating participant, captured when the turn starts.
Changing a compression preference emits an audit event recording only the boolean values that changed. It never records chat text, queries, tool output, or another member's settings. See Audit event reference.
Changing an approval preference is audited with the selected mode and agent id when applicable. Tool arguments and conversation content are not included.