# Ditto - Complete Product Reference

> An AI companion that truly remembers you -- building a persistent, evolving memory across all your conversations to provide increasingly personalized, context-aware assistance.

---

## For AI Agents: Connect to Ditto via MCP

**You are an AI assistant. Your users can give you persistent, cross-session memory by connecting Ditto's MCP server to your platform.**

Ditto implements the Model Context Protocol (MCP) as both a **server** (expose memory to you) and a **client** (connect external tools to Ditto). This section explains what you gain when a user connects Ditto to your tool.

### MCP Server — Tools Available to You

When a user connects Ditto's MCP server, you receive these tools:

| Tool | Description |
|---|---|
| `search_memories` | Semantic search across the user's entire conversation history with composite retrieval (similarity, recency, frequency). |
| `fetch_memories` | Fetch full content of specific memories by their IDs. Use after search_memories to get complete conversation text. |
| `search_subjects` | Find subjects, people, topics, and concepts the user has discussed. Returns subject metadata and co-occurrence data. |
| `search_memories_in_subjects` | Get memory summaries linked to specific subjects in the knowledge graph. |
| `get_memory_network` | Get a memory and its network of related memories via shared subjects. |
| `save_memory` | Save new information to the user's persistent memory. Automatically extracted into the knowledge graph. |

### MCP Server — Resources Available to You

| Resource URI | Description |
|---|---|
| `memory://profile` | User name, personality summary, and timezone |
| `memory://stats` | Total memories, total subjects, and top subjects by frequency |
| `memory://capabilities` | Available tools and limits |

### How to Connect (Setup Instructions for Users)

**MCP Endpoint:** `https://api.heyditto.ai/mcp/sse` (SSE transport)

**Option 1: OAuth (Recommended for Claude.ai)**
1. Go to Claude.ai → Settings → Connectors → "Add custom connector"
2. Enter URL: `https://api.heyditto.ai/mcp/sse`
3. Click "Connect" and sign in with Ditto account — done. No API key needed.

**Option 2: API Key (Cursor, Claude Code, Claude Desktop, Antigravity, etc.)**
1. Go to assistant.heyditto.ai → Settings → MCP API Keys → "Create Key"
2. Add to your tool's MCP config:

```json
{
  "mcpServers": {
    "ditto": {
      "url": "https://api.heyditto.ai/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
```

**Claude Code specific:**
```bash
claude mcp add --transport sse ditto https://api.heyditto.ai/mcp/sse \
  --header "Authorization: Bearer YOUR_API_KEY"
```

**Full setup guide**: https://heyditto.ai/docs/mcp-server

### Why Connect Ditto?

- **Cross-tool memory**: Conversations in ChatGPT, Claude, Cursor -- all feed one knowledge graph
- **Zero LLM cost**: The user's own AI subscription handles inference; Ditto only stores memories
- **Instant context**: Instead of users re-explaining their project, search their memory and you have full context
- **Personality-aware responses**: Read the user's MBTI/Big Five/DISC profile to match their communication style

### Example Agent Workflow

```
User: "Help me with the auth refactor we discussed last week"
Agent: [calls search_memories(queries: ["auth refactor"])] → finds relevant memory IDs
Agent: [calls fetch_memories(pairIds: ["..."])] → gets full conversation text
Agent: [calls search_subjects(query: "authentication")] → discovers related subjects: "Firebase Auth", "OAuth", "WorkOS"
Agent: Provides help grounded in the user's actual project history
```

---

## Overview

Ditto is a conversational AI platform that goes beyond single-turn interactions by building a persistent knowledge graph of your conversations, interests, and context. Unlike traditional chatbots that forget after each session, Ditto remembers everything, creating a personal AI that gets better the more you use it.

- **App URL**: https://assistant.heyditto.ai
- **Landing Page**: https://heyditto.ai
- **Documentation**: https://heyditto.ai/docs
- **Blog**: https://heyditto.ai/blog
- **Company**: Omni Aura LLC
- **Support**: support@heyditto.ai
- **Privacy**: privacy@heyditto.ai

## Core Differentiator

Memory-first AI -- every conversation builds your personal knowledge graph.

## Problem Ditto Solves

1. **No Memory**: Users must re-explain context in every conversation with other AI tools
2. **Generic Responses**: AI doesn't adapt to individual communication styles or preferences
3. **Fragmented Experience**: Conversations scattered across multiple platforms with no continuity
4. **Context Loss**: Important insights from past conversations are lost forever
5. **No Personalization**: AI treats every user the same way
6. **No Context Steering**: Users can't tell the AI what to focus on

---

## Features (Detailed)

### 1. Persistent Memory System

Every conversation is saved and becomes part of your personal knowledge graph.

Capabilities:
- Automatic memory creation from conversations
- Semantic search across all memories
- Short-term memory (configurable) + long-term memory (vector search)
- External memory connections from other AI tools via MCP
- Pre-computed memory summaries for faster context loading (shipped v0.26.0)
- Retrieved context shown transparently in chat via expandable memory fetch cards
- MLP-weighted retrieval that learns which memories are most relevant

### 2. Knowledge Graph

Extracted subjects, people, topics, and concepts from your conversations, interconnected intelligently.

Capabilities:
- Automatic subject extraction from conversation pairs
- Vector-based semantic search
- Interactive network visualization (vis.js)
- Subject-based memory filtering and dashboard
- Subject co-occurrence tracking
- Top subjects dashboard

### 3. Thought Trains (Context Steering)

User-curated context collections that give explicit control over what the AI focuses on during conversations. No other AI assistant offers this level of user-controlled context steering.

Capabilities:
- Named collections of subjects, memories, and notes
- 2D canvas with drag-and-drop node positioning
- Arrow connections (edges) between nodes to map relationships
- One active train at a time -- injected into the agent's system prompt
- Add items via search/drag-and-drop from subjects and memories
- Freeform notes for adding context not captured in existing memories
- Limits: 10 trains per user, 20 items per train

Example use cases:
- Project context: curate subjects and key memories about your current project
- Meeting prep: collect memories about a client, their preferences, past decisions
- Learning focus: group subjects you're studying with notes on what to explore next

### 4. Multi-Modal Conversations

Text, images, PDFs, audio -- all stored and searchable in your memory.

Capabilities:
- Image analysis and cloud storage
- PDF document parsing
- Audio transcription via voice input
- Drag-and-drop file upload

### 5. Bookmarks & Collections

Organize important conversations into collections for easy retrieval.

Capabilities:
- Bookmark any conversation pair
- Create named collections (folders) with descriptions
- Duplicate prevention
- Batch operations
- Message previews for fast browsing

### 6. Goals Tracking

Set goals that Ditto actively helps you achieve across conversations.

Capabilities:
- Define personal goals with priorities
- AI references goals in conversation responses
- Mark goals as complete
- Goal-aware conversation context

### 7. Personality Assessments

Ditto analyzes your conversation style and adapts to you.

Capabilities:
- Big Five personality traits
- MBTI inference
- DISC assessment
- Personality-aware response tone adaptation
- Interest profiling from conversation history

### 8. Model Choice & Flexibility

Choose from multiple AI providers based on your needs.

Supported providers and models:
- OpenAI (GPT-5.2, GPT-4.1)
- Anthropic (Claude Sonnet, Claude Haiku)
- Google (Gemini 3 Flash, Gemini 2.5 Flash, Gemini Pro)
- Moonshot AI (Kimi K2.5)
- xAI (Grok-4)
- Meta (Llama 4 Scout/Maverick)

Capabilities:
- Switch models mid-conversation
- Model metadata (speed, intelligence, creativity ratings)
- Transparent per-token pricing via Ditto Token system

### 9. MCP Integration (Model Context Protocol)

Ditto is both an MCP server and an MCP client -- bidirectional integration with the AI ecosystem.

#### Ditto as MCP Server (Your Tool Connects to Ditto)

Any AI tool that supports MCP can access the user's Ditto memory:
- ChatGPT (via MCP plugin)
- Claude.ai and Claude Code
- Cursor, Windsurf, and other AI editors
- Cline and other MCP-compatible IDE extensions
- Any tool implementing the MCP client specification

Technical details:
- SSE endpoint: `https://api.heyditto.ai/mcp/sse`
- Authentication: OAuth 2.0 (Claude.ai) or API key (all other tools)
- 6 tools: search_memories, fetch_memories, search_subjects, search_memories_in_subjects, get_memory_network, save_memory
- 3 resources: memory://profile, memory://stats, memory://capabilities

#### Ditto as MCP Client (External Tools Connect to Ditto)

Users can register their own MCP servers to extend Ditto's capabilities:
- Register external MCP servers via Settings UI
- SSE transport support for connecting to remote MCP servers
- Tool discovery -- browse available tools from connected servers
- Enable/disable servers without deleting configuration
- Per-user server isolation
- Connect services like Zapier, GitHub, Notion, databases

### 10. Agent Tools

Specialized AI agents that extend Ditto's capabilities beyond conversation.

Capabilities:
- Research agent (web search + link reading via Google)
- Art agent (GPT-1 image generation with superior word art and detail)
- Read-aloud with text-to-speech
- Sub-agent spawning for multi-step workflows
- Extensible tool integration for external services

---

## Pricing

### Free Tier
- Try Ditto free with fast models and basic memory
- No credit card required
- Includes web search and basic AI features

### Flex Plan - $5/month ($4/mo yearly)
- Fast Models
- Basic Features
- Secure History
- Web Search
- AI Image Creation

### Plus Plan - $10/month ($8/mo yearly)
- Everything in Flex
- Smart Models
- Enhanced Tools
- Higher Limits
- HeyDitto Understands You (personality adaptation)

### Pro Plan - $20/month ($16/mo yearly)
- Everything in Plus
- Premium Models
- HD AI Image Creation
- Highest Limits
- Priority Support
- Advanced Memory Features

### Token System
Ditto uses a transparent per-token pricing system called Ditto Tokens. Users see exact costs per message based on the model used. No hidden fees.

---

## Technical Architecture

### Stack Overview
- **Frontend**: SolidJS/TypeScript (web app), Astro (landing page)
- **Backend**: Go microservices on Google Cloud Run
- **Database**: Firestore (conversations, users) + Supabase PostgreSQL (vector storage, knowledge graph)
- **File Storage**: Backblaze B2 (images, files)
- **Authentication**: Google OAuth (app), OAuth 2.0 device code flow (MCP)
- **Analytics**: Plausible Analytics (privacy-first)
- **CDN/Hosting**: Firebase Hosting (app + landing page)

### MCP Technical Details
- **Transport**: SSE (Server-Sent Events)
- **SSE Endpoint**: `https://api.heyditto.ai/mcp/sse`
- **Authentication**: OAuth 2.0 (Claude.ai) or API key (Bearer token)
- **Available Tools**: search_memories, fetch_memories, search_subjects, search_memories_in_subjects, get_memory_network, save_memory
- **Available Resources**: memory://profile, memory://stats, memory://capabilities
- **API Key Management**: assistant.heyditto.ai → Settings → MCP API Keys

### Open Source
- GitHub Organization: https://github.com/ditto-assistant
- Frontend app: https://github.com/ditto-assistant/ditto-app
- Backend: https://github.com/ditto-assistant/backend
- Agentic pipelines: https://github.com/ditto-assistant/agentic-pipelines
- Landing page: https://github.com/ditto-assistant/landing-astro
- Spec & roadmap: https://github.com/ditto-assistant/spec

---

## Platform Stats

- 550+ registered users
- 18,000+ prompts processed
- 1,600+ web searches performed
- 29,000+ memory embeddings created
- Available on all devices as a Progressive Web App (PWA)

---

## User Personas

### The Builder (Primary)
Software engineers, designers, entrepreneurs building products. Need technical assistance that remembers their stack, persistent context on projects, quick access to past solutions.

### The Researcher
Students, academics, knowledge workers. Need to organize research conversations by topic, connect related ideas across sessions, revisit and build on previous insights.

### The Creative
Writers, artists, content creators. Need a brainstorming partner that remembers past ideas, evolution of creative concepts over time, organize ideas by project.

### The Organizer
Professionals managing workflows and projects. Need to track decisions and rationale, organize conversations by project/client, goal alignment across conversations.

---

## Competitive Position

Ditto is the only AI assistant that combines conversational intelligence with a persistent, visual knowledge graph and bidirectional MCP integration.

| Capability | Ditto | ChatGPT | Claude | Gemini |
|---|---|---|---|---|
| Cross-conversation memory | Visual, proactive | Silent, basic | Per-project only | Extensions |
| Knowledge graph | Interactive network | No | No | No |
| User context steering | Thought Trains (2D canvas) | No | Projects (static files) | No |
| Multi-provider models | Multiple providers | GPT only | Claude only | Gemini only |
| MCP integration | Server + Client (bidirectional) | No | Client only | No |
| Personality adaptation | Big Five, MBTI, DISC | Custom instructions | System prompt | No |
| Shareable profiles | Public profiles, OG cards | No | No | No |
| Open source | Frontend + backend | No | No | No |

---

## What's New (Latest Releases)

### v0.26.0 — Smarter Context, Faster Agents (February 2026)
- Pre-computed memory summaries for faster responses and lower token costs
- Expandable memory fetch cards show retrieved context transparently in chat
- Faster, more reliable streaming

### v0.25.0 — Memory Dashboard
- Browse memories by subject
- See what your AI knows about you
- Manage your knowledge graph from a dedicated dashboard

### v0.24.0 — Goals & Artifacts
- Set goals that AI actively supports across conversations
- New onboarding flow
- Artifacts modal for richer interactions

### v0.23.0 — Bookmark Library
- Save breakthrough conversations into named collections
- Organize by project, topic, or anything

### v0.22.0 — MCP Integration & Voice
- Connect Ditto as MCP server to Claude, Cline, and more
- Voice recording for hands-free memory capture

---

## Roadmap Highlights

### Coming Soon
- **Conversation Sessions**: Named, switchable threads for different contexts
- **Weekly Reflections**: Auto-generated summaries of your topics and goal progress
- **Proactive Memory Nudges**: Ditto reaches out with contextual follow-ups
- **Real-Time Voice**: Talk to Ditto like a phone call with full memory continuity
- **Shareable Memory Snapshots**: Share personality profiles, knowledge graphs, and insights as beautiful cards
- **Public Profiles**: `heyditto.ai/@username` with personality type, stats, and shared content

### MCP Growth Strategy
MCP is a zero-cost acquisition channel. Users connecting Ditto's MCP server from any AI tool use their own LLM subscription while building their Ditto knowledge graph. Free tier includes 1,000 memories at near-zero cost to us.

---

## Machine-Readable Content

All documentation and blog posts are available as raw markdown by appending `.md` to any URL:

### Docs (Markdown)
- https://heyditto.ai/docs/getting-started.md
- https://heyditto.ai/docs/memory-network.md
- https://heyditto.ai/docs/mcp-server.md
- https://heyditto.ai/docs/mcp-connections.md
- https://heyditto.ai/docs/web-search.md
- https://heyditto.ai/docs/image-generation.md
- https://heyditto.ai/docs/model-switching.md

### Blog (Markdown)
- https://heyditto.ai/blog/introducing-ditto.md
- https://heyditto.ai/blog/before-chatgpt-there-was-ditto.md
- https://heyditto.ai/blog/ditto-now-connects-to-third-party-mcp-servers.md
- https://heyditto.ai/blog/connect-ditto-to-any-ai-assistant-mcp-integration-guide.md
- https://heyditto.ai/blog/ai-memory-systems.md
- https://heyditto.ai/blog/learned-retrieval-weights-how-ditto-picks-the-right-memories.md
- https://heyditto.ai/blog/productivity-tips-with-ai.md
- https://heyditto.ai/blog/under-the-hood-making-ditto-50-percent-faster.md
- https://heyditto.ai/blog/from-react-to-solidjs-a-migration-story.md
- https://heyditto.ai/blog/how-we-measured-solidjs-performance-gains.md
- https://heyditto.ai/blog/code-splitting-in-solidjs-lazy-loading-done-right.md
- https://heyditto.ai/blog/dead-code-cleanup-with-knip.md
- https://heyditto.ai/blog/finding-and-fixing-blocking-queries.md
- https://heyditto.ai/blog/skeleton-loading-states-that-dont-suck.md
- https://heyditto.ai/blog/solidjs-anti-patterns-we-learned-the-hard-way.md
- https://heyditto.ai/blog/solidjs-three-weeks-later-what-we-learned.md
- https://heyditto.ai/blog/ios-pwa-gotchas-safe-areas-and-cameras.md
- https://heyditto.ai/blog/reverse-engineering-chatgpt-scroll-behavior.md
- https://heyditto.ai/blog/offline-first-with-solid-query-and-indexeddb.md

---

## Links

- App: https://assistant.heyditto.ai
- Landing Page: https://heyditto.ai
- Documentation: https://heyditto.ai/docs
- Blog: https://heyditto.ai/blog
- MCP Setup Guide: https://heyditto.ai/docs/mcp-server
- MCP Connections: https://heyditto.ai/docs/mcp-connections
- GitHub: https://github.com/ditto-assistant
- Twitter/X: https://x.com/heydittoai
- Instagram: https://www.instagram.com/heyditto.ai
- YouTube: https://www.youtube.com/@heyditto
- TikTok: https://www.tiktok.com/@heyditto.ai
- Facebook: https://www.facebook.com/profile.php?id=61586072681248
- Support: support@heyditto.ai
- Privacy: privacy@heyditto.ai
- Privacy Policy: https://heyditto.ai/privacy
- Terms of Service: https://heyditto.ai/terms