ModelTab

ModelTab direct mode. AI inference runs on the selected provider endpoint. Direct chat stores local settings/history and sends only chat context to the provider you choose. Optional Workspace Agent Mode can inspect a user-selected folder locally and share only visible read-only tool traces when you explicitly enable trace sharing.

Checking setup

ModelTab is validating provider, model, key, connection, and local recovery state.

Settings

Providers, memory, model controls, and key vault.

Provider

Active profile Provider preset

Base URL Model

Advanced provider options

Name Type API key No API key required for this OpenAI-compatible endpoint Extra headers JSON

Encrypted Key Vault

Passphrase

Static browser apps cannot hide keys from the browser runtime. Use session-only keys for maximum caution.

Instructions & Memory

System prompt Memory sent with every chat Chat context note

Prompt Library

Search prompts Saved prompt

Name Type

Tags Content Favorite this prompt

Type `/` in the composer for quick access. Variables like `{{topic}}` are filled before insertion.

Generation Controls

Temperature Top P Max output tokens Max input tokens Recent turns Streaming

JSON response mode where supported Gemini Google Search grounding Auto-trim request context to fit budget Include historical image attachments Extra request body JSON

Workspace Agent

Enable opt-in browser workspace mode

Send visible tool trace snippets with chat File path Search text Discovered files

Trace filter

Read-only tools run inside this browser tab against the folder you select. ModelTab never gets arbitrary disk access, never writes files, and never uploads raw files unless you attach them yourself.

Data

Data export excludes API keys. Full backup encrypts current session keys with the vault passphrase before download.

The service worker caches only the app shell. API requests and provider responses are not cached.

Connect a provider and start chatting.