The problem
Agents keep starting over.
Every step, agents re-read the same sources and redo the same math. Answers drift, and nobody can tell what a decision was based on.
About Kam
Build the agent, not the State plumbing.
Kam is building AI-native State infrastructure for Amazon Bedrock AgentCore: State management, event coordination, and a managed control plane on one foundation. Exact History and bounded State are implemented in source; managed lifecycle is the next step.
The approach
Define it once. Keep it current.
Say which results and numbers your agents need. Kam tracks what each one uses and updates only what a change touches.
The foundation
Exact revisions. Versioned membership.
Kam started with History: each record keeps a stable name, each change is a new version, and each list is fixed to one moment. That part is built and tested in code.
The origin
Built from hard lessons.
Founder John Yu built these pieces for an event-driven research system, where late fixes, retries, and recovery all had to agree. Kam turns that work into a general tool, with none of the old project’s logic.
Who does what
AgentCore runs agents. Kam keeps their State.
Kam
Kam owns State
- Normalized entities and exact versions
- Derived metrics and materialized views
- Change detection and dependency tracking
- History, corrections, and replay
Amazon Bedrock AgentCore
AgentCore runs the agents
- Runtime, sessions, and model calls
- Plans, decisions, and tool calls
- Gateway, identity, policy, and observability
- Memory for conversational and long-term recall
One project. A clear test.
Try Kam on one real project.
Tell us what your agents do and what keeps going wrong. We set what success looks like first.
Discuss an evaluationSee evaluation scopeNot launched yet. No production promise.