Amazon Bedrock
AWS's managed access to foundation models (Claude, Llama, Mistral, Nova and more) with retrieval, guardrails and fine-tuning around them.
Changes fast: check the latest updates too. Reviewed 17 Oct 2026.
Key ideas
- Converse API: one request format across models, with tool use and streaming.
- Knowledge Bases: managed ingestion, chunking and retrieval over S3 with OpenSearch, Aurora pgvector or S3 Vectors.
- Guardrails, prompt caching, batch inference, cross-region inference profiles.
- Provisioned throughput for guaranteed capacity; on-demand for variable load.
- IAM controls model access; use VPC endpoints for private traffic.
Common mistakes
- Region and quota limits; request quota increases before launch.