Verzeta

Frequently asked questions

Common questions about Verzeta Studio's intended behaviour. For things that don't work the way you expect, see Troubleshooting.

1. Where are my conversations stored?

All conversations, messages, attachments, tool results, polls, and activity log entries are stored in a local SQLite database file.

Location:

The folder also holds project documents, canvas files, skills and logs. On Linux, downloaded GGUF models are stored there too; on Windows they are in %LOCALAPPDATA%\Verzeta\verzeta-studio\.

Pro-tip: To back up your Verzeta data, close the app and copy the entire data folder. To move to a new computer, restore the folder there before opening the app, then enter your API keys again (they do not transfer; see question 6).

Note: API keys are not in this folder. See question 6.

2. Are my prompts or conversations sent anywhere I haven't explicitly opted into?

No. Verzeta has no telemetry, analytics, usage reporting or cloud sync.

Your messages go only to places you set up:

There are no analytics, no developer dashboards, and no Verzeta-operated servers.

Note: When you use a cloud provider (OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek), that provider receives the message content per their terms of service. This is the same as using any other API client. Verzeta sends the provider what the model needs to answer: the system prompt, recent messages, any summary or retrieved context, and tool definitions. It sends nothing else.

3. How do I switch the model used for one conversation?

Open the conversation. Open Chat Settings (the gear button in the chat header) and change the Model section.

The change applies to the next message you send. Previous messages in the conversation are not re-run.

If you want different agents within the same group conversation to use different models, see the per-member override section of Multi-agent teams. That feature lets you assign one provider to @Alice and another to @Clark in the same chat.

4. Can I run Verzeta completely offline?

Yes. There are three ways to run models locally, and you can mix them:

What still requires connectivity even in "offline" mode:

Note: RAG works offline when Settings → Providers → Embeddings (RAG) uses a local model (Local AI edition) or an embeddings server on your own network.

5. What's the difference between a regular conversation and a group chat?

Aspect Regular conversation Group chat
Number of agents One Two or more
Speaker on each turn Always the same assistant Whichever member is dispatched (cascade decides)
@mention routing No: @mentions are plain text Yes: @mentions dispatch to the named member
Member overrides One model per conversation Per-member provider, model and tools
Polls Available but rarely used Often used for coordinated decisions
Created via New Chat New Group Chat
Best for Solo Q&A, drafting with one assistant, quick lookup Multi-agent collaboration, team workflows, coordinated decisions

Both kinds support the same per-conversation settings (system prompt, temperature, RAG, tools, etc.). The difference is whether several named members take part.

For the deep version, see Multi-agent teams.

6. Where are my settings and API keys stored?

Settings and API keys are stored separately from your conversations:

API keys are scrambled so they cannot be read at a glance, but they are not encrypted. Anyone who can read your user account's files can recover them. They only work on the computer where you entered them, so after moving to a new computer you must enter them again.

7. How do I remove all my data?

Deleting the AppImage or the Windows folder does not delete your data. To remove everything, close Verzeta Studio and delete:

To delete only your conversations, use Settings → Storage → Clear All Conversations. This cannot be undone.

8. Can I switch between the Standard and Local AI editions?

Yes. Both editions use the same data folder and settings, so your conversations and providers carry over. The built-in llama.cpp engine and local embeddings only work in the Local AI edition.

9. Is there a list of keyboard shortcuts?

Yes. See Keyboard shortcuts.

10. Why did my agent team stop?

A team stops when it has nothing left to do, when a round made no progress, or when it has used up its automatic rounds (6 by default). The chat always posts a notice saying which. Reply with anything to start it again, or raise Team autonomy: rounds before pausing in Chat Settings → Agent. A single agent also pauses after 25 tool calls in one turn (Tool steps per turn). See Turn limits, rounds, and pauses.

11. Are agent shell commands sandboxed?

No. The shell tool only runs programs on your allow-list and refuses destructive patterns, but the commands themselves run with your user's permissions. Only canvas Run uses a sandbox (bubblewrap, on Linux). See Agent execution and permissions.

Verzeta Studio guides