FAQ

Backboard is a unified AI infrastructure platform that provides access to 17,000+ LLMs, built-in RAG, and the world’s smartest memory system—all through a single API. It enables developers to build stateful, context-aware AI applications without managing complex infrastructure.

Backboard uses a sophisticated hybrid storage system to manage and retrieve memories for AI assistants. When you send a message with memory="Auto", the system automatically extracts relevant facts and preferences, storing them across optimized data stores.

When your assistant needs to access memories, Backboard conducts a comprehensive search and retrieves relevant information that can be seamlessly integrated into responses—enabling personalized, context-aware interactions across conversations.

Backboard’s memory implementation offers several advantages over traditional Retrieval-Augmented Generation:

• Entity Relationships: Understands and relates entities across different interactions, not just static documents
• Contextual Continuity: Retains information across sessions, maintaining continuity in long-term engagements
• Adaptive Learning: Improves personalization based on user interactions and feedback
• Dynamic Updates: Updates memory in real-time with new information, unlike static RAG systems

• Personalized Learning Assistants: Remember user preferences, strengths, weaknesses, and progress
• Customer Support AI Agents: Retain context from previous interactions for better assistance
• Healthcare Assistants: Track patient history, medication schedules, and treatment plans
• Virtual Companions: Build deeper relationships by remembering personal details and past conversations
• Productivity Tools: Remember user habits, frequently used documents, and task history
• Gaming AI: Create immersive experiences by remembering player choices and progress

1. Sign up at app.backboard.io
2. Navigate to Settings → API Keys to create your API key
3. Install the SDK (pip install backboard-sdk or npm install backboard-sdk)
4. Follow our Quickstart Guide to send your first message

No. New accounts receive $5 in free credits (valid for 30 days) to explore the platform. You can start building immediately without entering payment information.

Backboard provides official SDKs for:

• Python: pip install backboard-sdk
• JavaScript/TypeScript: npm install backboard-sdk

You can also integrate directly via our REST API using any HTTP client.

Simply set memory="Auto" when sending a message. Backboard automatically extracts and stores relevant information, then retrieves it when needed in future conversations.

response = await client.send_message(
    "My name is Sarah and I work at Google.",
    memory="Auto",
)

Backboard uses a classification system to determine which content should be stored as memories. Not all text will generate memories. Memories are typically not created for:

• Definitional questions (e.g., “What is machine learning?“)
• General concept explanations without personal context
• Technical definitions and theoretical explanations
• Abstract or theoretical content

Best practices for memory extraction:

• Include temporal markers (when events occurred)
• Add personal context or experiences
• Frame information in terms of real-world applications
• Include specific examples rather than general definitions

Yes. You can view, enable, disable, or delete individual memories through the dashboard or via the Memory API endpoints. This gives you granular control over what your assistant remembers.

Yes. Memories are stored at the assistant level, not the thread level. This means information saved in one conversation thread is accessible in all other threads for the same assistant.

Backboard provides access to over 17,000 models from leading providers including:

• OpenAI (GPT-4o, GPT-4)
• Anthropic (Claude 4, Claude 3.5)
• Google (Gemini Pro, Gemini Ultra)
• X.ai (Grok)
• Cohere (Command R+)
• Featherless (open-source models)
• OpenRouter (frontier and free models)
• Cerebras (optimized for speed)
• AWS Bedrock

Yes. Backboard’s stateful thread management supports on-demand model switching across all 17,000+ models. You can specify different models for different messages within the same thread.

No. Backboard provides unified access to all models through a single API key. You don’t need to manage separate credentials for each provider.

Backboard supports common document formats including PDF, TXT, and other text-based files. Documents are automatically indexed and made searchable for RAG queries.

Backboard’s agentic RAG system uses hybrid search combining BM25 and vector retrieval. It automatically chunks documents based on the model’s context window, achieving low-latency retrieval while seamlessly scaling to accommodate additional documents.

Yes. While Backboard provides built-in RAG capabilities, you can integrate your own vector database and retrieval system if needed.

Backboard uses a credit-based system. You pay for the tokens and resources you consume. New accounts start with $5 in free credits (valid for 30 days). Visit app.backboard.io for current pricing details.

Yes. Backboard supports multi-user accounts, allowing you to share credits across multiple users under one billing profile.

• Documentation: Browse our docs for guides and API references
• Status Page: Check status.backboard.io for service status
• Contact: Reach out via the dashboard for support inquiries

You can monitor real-time service status at backboard.statusgator.app. We recommend implementing appropriate error handling and retry logic in your applications.