AWS Bedrock + Enhanced File Upload

What's New

AWS Bedrock models are now available through the unified Backboard API with full memory, configuration, and monitoring support. This brings enterprise-grade AWS infrastructure directly into your Backboard workflows.

New Models

• Anthropic Claude 4.5 Haiku & Sonnet — Fast, affordable models with 200K context for production workloads.

• Claude 4 Sonnet & Opus — Top-tier reasoning with extended context windows.

• Meta Llama 4 Maverick — 1M context window for massive document processing.

• Meta Llama 4 Scout (128K) & Llama 3.3 70B — Versatile open models for varied use cases.

• DeepSeek R1 — Advanced reasoning for complex multi-step tasks.

• Mistral Pixtral Large — Multimodal vision + text for image understanding.

Enhanced File Upload Limits

We've significantly expanded file upload capabilities across all tiers:

• Assistant level: 20 files per assistant, 5M tokens per file, 200 MB maximum file size.

• Thread level: 20 files per thread, 140 total files across assistants and attachments.

• Message attachments: 10 per message, 100 per thread, 1M tokens each, 10 MB max.

Improved Error Handling

• Detailed error messages now include token usage info for easier debugging.

• File validation errors surface before processing begins — no wasted compute.

• Rate limit responses include retry-after headers for all Bedrock models.

How to Use It

Set provider: "aws-bedrock" in your assistant configuration, or select any Bedrock model from the Model Library. All existing memory, RAG, and routing features work automatically with the new models.