Profiles & AI Providers · ZeroChat Help

1. Connection Profile Management

ZeroChat lets you define multiple connection profiles. A profile encapsulates the API protocol type, server endpoint URL, authentication key (optional for local servers), and selected model name.

To manage profiles, click the active profile name in the header. The selector lets you create and export profiles, close the menu, and edit or delete each listed profile. Selecting one makes it active.

Instant switching: You can switch models or providers mid-conversation without losing context. Simply open the profile dropdown and select another one.

The profile selector lets you Export all editable definitions. It downloads an encrypted HTML file: open it to start importing into ZeroChat. After confirmation, new profiles are added and matching identifiers are updated without changing the Mirror profile.

API keys are encrypted in the browser's local profile storage and remain encrypted inside the backup, which adds a second encryption layer. The Lock changes checkbox, located in the editor header, marks a profile as non-editable when it is loaded again; the lock protects all profile settings. Changes are saved using the Save button in the modal header.

From Settings → Encryption password, you can protect API keys with your own password. ZeroChat never stores the password: it temporarily keeps its hash in browser storage for 24 hours so unlocked access is shared across tabs. You will need to enter it again afterwards. Use default key requests the current key and re-encrypts all API keys with ZeroChat’s built-in key.

2. Supported Providers & Protocols

Interface Type	Common Servers & Services	Typical Endpoint	API Key
OpenAI / Compatible	LM Studio, LocalAI, vLLM, official OpenAI	`http://localhost:1234/v1`	Optional on local / Required on OpenAI
Ollama	Local Ollama server	`http://localhost:11434`	Not required
OpenRouter	OpenRouter multi-model gateway	`https://openrouter.ai/api/v1`	Required (`sk-or-...`)
Anthropic Claude	Official Anthropic API	`https://api.anthropic.com/v1`	Required (`sk-ant-...`)
Google Gemini	Google AI Studio	Gemini OpenAI-compatible endpoint	Required
WebLLM	In-browser direct (WebGPU)	`webllm://local` (automated)	Not applicable (100% local)
Mirror (Espejo)	Internal offline testing	No endpoint (simulated mode)	Not applicable

3. Server Inspection (Query Button) & Validation

When creating or editing a profile, ZeroChat provides a Query button next to the URL input. This inspects the server before saving:

Model discovery: Fetches the available models catalog and populates the dropdown automatically.
Capability inspection: Verifies if the model supports vision, function calling (tools), native reasoning, and advertised context capacity.
Save protection: For remote APIs or local HTTP endpoints, ZeroChat requires a successful Query before saving to ensure the connection works properly.

Fast save on WebLLM: When selecting an already downloaded WebLLM model from local cache, the Query button is optional, allowing you to save the profile immediately.

4. Special Configuration for Ollama (CORS)

When connecting to Ollama, browsers will block requests due to CORS (Cross-Origin Resource Sharing) unless Ollama is started with cross-origin access enabled:

# On Linux / macOS (terminal)
OLLAMA_ORIGINS="*" ollama serve

# On Windows (PowerShell)
$env:OLLAMA_ORIGINS="*"
ollama serve

If running Ollama as a systemd service on Linux, add Environment="OLLAMA_ORIGINS=*" under the [Service] section.

5. The Built-in "Mirror" Profile

ZeroChat includes a read-only Mirror (Espejo) profile. It is ideal for UI testing and troubleshooting:

Makes zero network calls and needs no local server running.
Returns an immediate formatted preview of the normalized OpenAI-compatible payload.
Serves as a reliable fallback whenever an active profile is deleted or inaccessible.