Built on Amazon Web Services
Pika runs on proven AWS services so you inherit scalability, security, and operational maturity from day one.
Architecture essentials
- Amazon Bedrock: Executes the LLM and inline agents (e.g., Anthropic Claude) with native function calling.
- AWS Lambda (Function URL): Stateless compute for chat orchestration and streamed responses.
- Amazon API Gateway: REST APIs for sessions, messages, titles, and suggestions.
- Amazon DynamoDB: Durable storage for chat sessions, messages, users, and dynamic agent/tool definitions.
- Amazon S3: Stores generated insights as JSON; optional asset storage.
- Amazon OpenSearch Service: Indexed sessions, feedback, and insights for search and analysis.
- Amazon EventBridge: Schedules the insights runner and other background jobs.
- AWS IAM: Least‑privilege access; tools are tagged (e.g.,
agent-tool
) for safe Bedrock invocation.
What this gives you
- Operational simplicity: No servers to manage; CDK sets up infra and wiring.
- Elastic scale: Spikes in chat traffic are handled automatically.
- Security by default: IAM, VPC options, and encrypted cookies minimize risk.
- Clear observability: Traces, usage, and costs are captured per message.
Deployment
You deploy with AWS CDK: separate frontend and backend stacks, plus hooks for adding your own IAM policies and resources.
Replace mock auth before production
The default dev auth is intentionally insecure. Implement your provider and assign user types and roles before going live.
On this page