Ninth Wave deployed the multi-agent assistant Compass for bank onboarding on Amazon Bedrock AgentCore
Ninth Wave built the multi-agent assistant Compass to onboard banking API partners into an open finance network. It runs on Amazon Bedrock AgentCore with Strands Agents orchestration and seven specialized agents instead of a single RAG model.
Ninth Wave, a provider of connections between banks and third-party applications within open finance (a network for sharing banking data with customer consent through standardized APIs), built the Compass assistant for onboarding banking partners. The tool runs on Amazon Bedrock AgentCore and addresses the problem that each bank offers APIs with its own field names and formats that differ from the FDX (Financial Data Exchange) standard — according to the company, validating and mapping such APIs previously required weeks of work by specialists using emails and spreadsheets.
According to the source, Compass is built on a multi-agent architecture with the Strands Agents orchestration framework. The main agent classifies the intent of a request and routes it to one of seven specialized agents, each of which works with its own context window and instructions for a specific task (e.g. field mapping, documentation search, readiness analysis). The company chose this approach after considering alternatives — self-hosted models on Amazon EC2 (full control, but higher overhead) and a single RAG agent (simpler, but less accurate across task types, according to the company).
The tool serves as a self-service portal for banking engineers, integration teams at aggregators (e.g. Plaid, Finicity, MX), and the Ninth Wave onboarding team. Security is provided by a layer of AWS WAF and CloudFront at the network edge, authentication through OAuth2/OIDC with multi-factor authentication, AWS Secrets Manager with environment-level KMS keys, and strictly segregated access by tenant through separate indexes in OpenSearch and prefixes in S3. AI workloads run separately from the application layer through a cross-account IAM role.
The source article from AWS is incomplete — it lacks a description of the rest of the architecture and any deployment results. You can find details in the source article.
Why it matters
The case shows a concrete production deployment of a multi-agent architecture in a regulated industry (financial services), where isolating data from individual banks and ensuring access is auditable are essential. For teams designing similar agent systems, it provides evidence that dividing tasks among specialized agents with their own context can be more accurate than a single RAG agent sharing prompt space across query types.
Two audiences, two different impacts
What this means
For individuals
For developers of agent systems, this is a concrete reference design showing how to divide tasks among specialized agents instead of using a single RAG agent with a shared context window, and which security layers (tenant-scoped grounding, cross-account isolation of AI workloads) accompany this approach.
For a business
For financial institutions and fintech companies working on open finance integrations, this is a documented example of deploying a multi-agent AI system in a regulated environment with banking security requirements (IAM, KMS, per-tenant data isolation) that, according to Ninth Wave, speeds up the onboarding process for banking API partners.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.