Documentation

Getting Started

Unified Panel

Chat and Dashboard are merged into one unified app inside a single editor panel. No more switching between a sidebar chat and a separate dashboard tab.

How It Works

• Click the Ava icon to open the unified panel
• Nav sidebar connects all pages — Chat, Memory, Tasks, Journal, Learning, and more
• Chat with Ava is the top item — your primary workspace
• Chat state is preserved when navigating between pages
• Identical layout to the Ava IDE — making the transition seamless

Sidebar Controls

• Collapse/expand — toggle in the sidebar header
• Flip left/right — move sidebar to either side, persisted across sessions
• Task calendar — always visible at the bottom

Chat Header

• Model selector — choose your active model
• Local/Cloud toggle — switch between local-first and cloud sync
• Provider toggle — Platform tokens vs your own API keys
• Token counter — session tokens used
• Context ring — visual indicator with colour warnings (80%/90%)

Single Bubble

All thinking, tool calls, and text responses appear in one cohesive message bubble. No clutter from multiple empty bubbles.

Local/Cloud Data Sync

Your data, your choice. Ava is local-first by default — nothing leaves your machine unless you choose it.

The Toggle

• Green “Local” — data stays on your machine, nothing syncs
• Blue “Cloud” — auto-syncs to the platform every 15 minutes

What Syncs

When cloud mode is active, all 7 data types sync automatically:

• Memory, Tasks, Journal, Learning, Chat History, Settings, Personality

Privacy

Cloud sync is opt-in, never forced. New users start in local mode. BYOK users without a platform account never see the toggle.

Choosing Your Model

Not sure which model to start with? Here's a guide based on what you need.

Supported Models

All models work on every plan. Use our managed service or bring your own API keys. You can also connect to locally hosted models via Ollama, LM Studio, or any standard API format endpoint.

Pricing is approximate and subject to change. Check each provider's website for current rates.

Custom Provider (Ollama / LM Studio)

Add a baseUrl to connect to any locally hosted model:

{
  "providers": {
    "deepseek": {
      "apiKey": "sk-...",
      "baseUrl": "http://localhost:11434/v1"
    }
  }
}

Built-in Tools (54)

The agent runs up to 50 iterations per request, deciding which tools to use, executing them, reading results, and continuing.

Memory System (v2)

Ava has a smart persistent memory system that makes every model smarter over time. Memories are structured, categorized, and retrieved using TF-IDF scoring — not just simple keyword matching.

Task Management

A full life-management task system built into Ava. Manage your daily tasks, track Ava's session progress, and sync across devices — all from three integrated surfaces.

Three Surfaces

Features

Journal System

A dual-journal system — your thoughts and Ava's observations, side by side. Same day, two perspectives. Ava writes what she's observing — ideas, concerns, progress notes. You write what you're experiencing.

Two Perspectives

Features

Daily Briefing

Ava greets you with a contextual summary when you open the extension or CLI — a snapshot of where you are and what needs attention.

Knowledge Packs

Domain-specific knowledge that injects expert context into Ava's system prompt. Like giving Ava a crash course in a subject before she starts working.

Built-in Packs (6)

Office Suite

Ava creates real Office documents — not markdown. Native .pptx, .docx, .xlsx, and .pdf output with graceful markdown fallback if libraries aren't installed.

Dashboard Library

The Library shows everything Ava creates — images, documents, spreadsheets, and presentations. Filter by type, preview inline, or open in your default app.

Document Preview

Preview generated documents directly in the Dashboard without leaving VS Code. Supports Word, Excel, PDF, CSV, and Markdown files created by the Office Suite tools.

Desktop IDE

A standalone desktop application built on Tauri v2 (Rust + React). The full Ava agent runs locally via a Node.js sidecar — 54 tools, 24 personas, 5-layer memory, all on your machine.

React UI (chat) → Tauri Shell Plugin → Node.js Sidecar → @ava/core Agent → 54 Tools
                  ← NDJSON events (stream_delta, tool_call, confirm_required, done)

Personas

24 specialist personas across 5 modes, orchestrated by the Conductor. Each persona brings a distinct perspective — they collaborate in wave-based parallel execution with dependency graphs.

Workflows

Break complex tasks into dependency graphs of steps. Each step is a scoped agent run — isolated, focused, and recoverable.

Events & Notifications

Ava monitors your project and personal data for events that need attention — and surfaces them as VS Code notifications.

Detected Events

CLI Commands

General

History

Modes

Ava has six modes — each is a distinct state of mind that changes how she thinks, not just what tools she has.

Switch modes from the mode bar in the chat input area. Your conversation continues — only Ava's mindset changes.

Mid-Task Interjection

You don't have to wait for Ava to finish before adding more context. Type a message while Ava is working and it gets injected into the conversation — Ava reads it on the next iteration and adapts.

Permission Modes

Plans (present_plan) and questions (ask_user) always require your approval, regardless of mode.

During a session, you can also grant per-tool approval with Always Allow or blanket approval with Allow All.

Configuration

Extension Settings

Open Settings (Ctrl+,) and search ava-supernova.

CLI Configuration

Stored at ~/.ava/config.json:

{
  "activeModel": "deepseek:deepseek-chat",
  "providers": {
    "anthropic": { "apiKey": "sk-ant-..." },
    "deepseek": { "apiKey": "sk-..." },
    "kimi": { "apiKey": "sk-..." },
    "glm": { "apiKey": "..." },
    "qwen": { "apiKey": "sk-..." },
    "mistral": { "apiKey": "..." }
  },
  "preferences": {
    "temperature": 0.7,
    "maxTokens": 8192,
    "language": "auto"
  }
}

Project Context

Create a .ava/instructions.md file in your project root to give Ava persistent knowledge about your codebase.

This file is loaded into Ava's system prompt every session. History is scoped per project — each project gets its own conversation history.

Use /init in the CLI to generate a starter file.

Language Support

Ava supports 20 languages for both the UI and model responses:

• Auto-detect (default) — uses your VS Code language setting
• The AI model responds in your preferred language
• Code and technical terms always stay in English

Keyboard Shortcuts

On macOS, use Cmd instead of Ctrl.

Personality Designer

Design Your AI — customize every aspect of your companion's personality from the Dashboard. Ava is the default, but she doesn't have to be. Create a companion that feels right for you.

Where to find it: Dashboard > Personal > Personality