Assistant

The Assistant is your AI-powered pair programmer. It now runs as an independent, standalone application (The Hub) that connects to the Suite and other IDEs via the Model Context Protocol (MCP). It is context-aware, tool-enabled, and fully customizable. While "Ollie" is the default persona, you can create and manage an entire team of specialized agents.

The Brain (Providers)

You can choose the LLM that powers the Assistant. Switch between them instantly from the connection tab:

Cloud High-IQ: Connect to providers like OpenAI (GPT-4), Anthropic (Claude 3.5), or Gemini for complex architectural tasks. (Requires API Key)
Local & Private: Use Ollama, LM Studio, or LocalAI to run models entirely on your machine. Your code never leaves your computer.

Agent Manager

The Assistant is designed to be modular. You are not limited to one "bot".

Create Custom Agents: Define new agents with specific system prompts (e.g., "Senior React Dev", "SQL Optimizer").
Share & Sync: Export your best agents as JSON files to share with your team or import agents created by others.
Cartridges (.octdg): Portable, self-contained agent packages. A cartridge bundles an agent's identity, a Brain (JavaScript logic), and its own dependencies. Install one and it "just works." Ollie ships with two default cartridges:
- Ollie 🤖 — A zero-LLM deterministic assistant with 15+ built-in skills. Works instantly offline: file operations, app generation, project stats, web search, calculator, dice rolls, timers, knowledge base, and voice-driven OS control ("click on Submit", "type hello", "scroll down"). Think Alexa-era smart assistant — fast, reliable, no API key needed.
- Mango 🥭 — An LLM-powered AI companion inspired by the movie Her. Warm, curious, and conversational — Mango connects to your choice of LLM provider and can reason about complex tasks, control your OS, browse the web, and have natural multi-turn conversations. Requires a model to be configured (local or cloud).
Agent Delegation: Agents can call other agents as tools. This enables multi-agent orchestration — for example, a "Tech Lead" agent that delegates sub-tasks to a "Frontend Dev" and a "Backend Dev" agent automatically.

Context Intelligence

The Assistant uses a sophisticated "Prompt Intelligence" system to manage what the AI sees:

Smart Context ({{SMART_CONTEXT}}): Automatically selects relevant files based on your query using keyword matching and recency.
Context Rules: Configure behavior like "Auto-load READMEs" or "Allow Binary Files" (images/videos) per agent.
Manual Control: Drag & drop files, or use @ mentions to manually pin context.
Visual Map: View the "Directory Tree" preview to see exactly what structural knowledge the agent has.

Tools & Permissions

Agents can be granted Tools to perform actions. You have granular control over permissions for each tool:

Filesystem

read_file — Read file contents
create_file — Create new files
modify_file — Edit existing files (with diff view)
create_folder — Create directories
delete — Remove files or folders
list_folder — List directory contents

Search & Navigation

search_files — Find files by name
grep_search — Search file contents
display_file_ide — Open a file in the editor
web_search — Search the web

Execution

run_command — Execute terminal commands
generate_interactive_app — Create .iapp tools
open_app — Launch interactive apps

System Control

take_screenshot — Capture the screen
mouse_control — Move and click the mouse
keyboard_control — Type and press keys
click_text — Find text on screen via OCR and click it

Permission Levels: Set each tool to "Always Allow", "Ask for Approval" (Human-in-the-loop), or "Disable". View the full mapping in the "Prompt Intelligence" tab.

MCP Integration

Ollie supports the Model Context Protocol (MCP), a universal standard for AI tool plugins. This means you can extend your agents with community-built tools:

Connect MCP Servers: Configure external MCP servers (e.g., github_mcp, postgres_mcp) from the connection settings.
Transparent Access: MCP tools appear alongside built-in tools. Because the Assistant itself runs via MCP, it effortlessly bridges the gap between your remote plugins and the local IDE.
Three Sources: Tools come from three layers: Core (shipped with Ollie), User (your custom scripts in ~/.ollie-assistant/tools/), and MCP (plugins and the Suite itself).

Voice Mode

Code hands-free with full voice support.

Talk to Assistant: Click the Microphone icon to speak.
Hear Responses: Enable the Speaker icon for Text-to-Speech (TTS) output.

Autonomous Agents (Mango)

For advanced automation, use Mango — a conversational AI companion inspired by the movie Her. Unlike traditional assistants, Mango has personality: warm, curious, and genuinely interested in helping you.

Natural Conversation: Structured multi-turn dialogue with proper context memory — Mango remembers what you've been talking about, not just the last message.
Vision: Can "see" your screen via screenshots to debug UI issues, read error dialogs, or navigate visual interfaces.
Voice-Driven OS Control: Say "click on Submit" and Mango will screenshot your screen, find the text via OCR (click_text tool), and click it — hands-free computer navigation.
Keyboard Dictation: Say "type hello world" and Mango types it directly into the active window.
Multi-Provider: Connect Mango to any LLM — Ollama (local), OpenAI, Anthropic, Gemini, Groq, and 8+ more providers. Configure via show settings.
Independent Loop: Runs in a self-governed loop to solve multi-step tasks without constant prompting.