Settings

Verzeta has two kinds of settings: conversation settings, which control how one chat behaves on the desktop, and extension settings, which control the VS Code extension itself.

Conversation settings

Click the gear in the chat header. The panel is titled Chat settings. Most of these settings are stored on the desktop and apply to the conversation in every paired client (VS Code, Android and so on). Command execution is the exception: it is stored on this device.

Model

Shows the current provider and model. Click Change (or the model name in the chat header) to pick another one. The list contains the models set up on the desktop: local Ollama models, or cloud providers (OpenAI, Anthropic, Gemini and others) with your own API keys.

System prompt

Set or edit the system prompt, which is sent at the start of every request in this conversation. Leave it blank to use the agent's default.

Parameters and sampling

Setting	What it does
Temperature	How random the model's output is
Max tokens (-1 = Auto)	The most tokens the model may write per reply. -1 uses the model's own default.
Context window	How much history the host includes in each request
Use app-recommended sampling	Use the host's tuned sampling profile for the active model. Turn it off to set the values below yourself.
Top K, Top P, Repeat penalty, Presence penalty, Frequency penalty	Sampling controls. Each accepts -1 for Auto.

Conversation memory

Auto-compact long chats summarises older messages as the context window fills, so long conversations keep fitting in the model's context. Compact every N turns (0 = off) also compacts on a fixed schedule.

Agent behaviour

Setting	What it does
Agent pattern	How the agent works through a request: Direct, ReAct, Planner, Router, Multi-agent or Memory
Streaming	Show replies as they are generated
Thinking	Ask models that support it to reason before answering. The Thinking toggle next to the send button does the same.
Require confirmation	Pause the agent before it runs destructive tools
Tools enabled	Let agents use the host's tools. This applies to every conversation on the host.
Describe tools in system prompt	Add long tool descriptions to the system prompt. The panel recommends leaving it off.
RAG enabled	Add retrieved context to the prompt, for this conversation

When Thinking is on and the model reasons, a collapsed Reasoning section appears above the reply. Models without a reasoning pass ignore the setting.

To limit which tools a member of a project can use, set that up on the desktop.

Other sections

Command execution: whether agents may run commands in your shared workspace on this device. See Workspace mount.
Heartbeat: Auto-surface heartbeat reports.
Preferred skills: choose skills for this chat. Turn on Override parent folder to use your own choice instead of the folder's.

Extension settings (`verzeta.*`)

These settings are in your VS Code settings (settings.json) and control the extension itself. Open them with File > Preferences > Settings (Ctrl+, / Cmd+,) and search for "Verzeta".

Hosts

Setting	Default	Description
`verzeta.hosts`	`[]`	The paired Verzeta Studio desktops, each with an `id`, `name`, `url` and optional `tlsCertSha256` (the TLS pin). Use Verzeta: Add Host and Verzeta: Remove Host… instead of editing it.
`verzeta.defaultHostId`	(empty)	The `id` of the host to use at startup. When empty, the first paired host is used.
`verzeta.connectOnStartup`	`true`	Connect to the default host when VS Code starts.

Interface

Setting	Default	Description
`verzeta.initialTab`	`home`	Which tab the Verzeta view opens on: `home`, `chat` or `settings`.
`verzeta.log.level`	`info`	How much the Verzeta output channel logs: `error`, `warn`, `info` or `debug`. Set it to `debug` when troubleshooting connection problems.

CodeLens

Setting	Default	Description
`verzeta.codeLens.enabled`	`false`	Show Verzeta CodeLens links (Explain, Add tests, Refactor) above functions and classes. Off by default.

Workspace Mount

These settings control how agents on the desktop read and write files in your shared workspace. All verzeta.workspaceMount.* settings, including autoRegister and allowlist below, apply from your user settings only. A workspace's .vscode/settings.json cannot change them.

Setting	Default	Description
`verzeta.workspaceMount.defaultTier`	`ask`	The permission tier pre-selected when you share a workspace: `ask` or `smart`. Also used when you share from the in-chat banner, the share prompt or automatically. Any other value, including `bypass`, is treated as `ask`; choose Bypass in the tier picker.
`verzeta.workspaceMount.blocklist`	`[]`	Extra path globs that agents may not list, read or write. The built-in blocklist (credentials, keys, `.env` files, `.git`, `node_modules`, build output) always applies and cannot be removed.

Smart Mode options

These rules run on every write. A write whose extension is not on the list is refused in every tier. In Smart, a write over the size or line limits asks for confirmation.

Setting	Default	Description
`verzeta.workspaceMount.smartMode.extensionAllowlist`	(long list of code and text types)	File extensions agents may write to. Writes to other extensions are refused in every tier.
`verzeta.workspaceMount.smartMode.maxDiffLines`	`500`	In Smart, a write that changes more lines than this asks first.
`verzeta.workspaceMount.smartMode.maxFileBytes`	`262144` (256 KiB)	In Smart, a write larger than this asks first.

Setting	Default	Description
`verzeta.reconnect.backoffMs`	`1000`	Delay in milliseconds before the one automatic reconnect attempt after a dropped connection (250 to 60000).
`verzeta.workspaceMount.autoRegister`	`true`	Share the workspace automatically on the first message in a project room that has no mount, when exactly one workspace folder is open. See Workspace mount.
`verzeta.workspaceMount.allowlist`	`[]`	When not empty, agents can only list, read and write paths matching one of these globs.