Skip to main content

Kam State · for Amazon Bedrock AgentCore

Build more. Manage less state.

Shared State, exact History, version checks, and recovery, without building them yourself.

Explore the productDiscuss an evaluation

Prelaunch. No hosted service yet.

Loading the example. Open Explore

What Kam does today

  • Exact

    Ask for a version and get that one, or a clear no.

  • Kept

    Fixes add versions. Nothing is overwritten.

  • Checked

    Access is checked on every read. A browser can’t grant itself access.

  • Capped

    Each read has a fixed work limit, so it won’t grow with your data.

Built into the code. Speed is measured after live testing.

One agent turn
Explore product layers

Before Kam, and after

Most agent headaches have one root: shared results no one keeps. It shows up three ways.

Root problem: cause → costWith Kam: the whole class removed

Copies drift.

Each agent keeps its own copy.Agents act on different answers.

  • Stale reads
  • Silent overwrites
  • Double counting

Every agent can ask for an exact version. A fix adds a version. A retry counts once.

Inputs go missing.

Inputs change or vanish.No one can show what an agent saw.

  • Mystery inputs
  • Lost evidence
  • Unrecorded sources

Recorded sources are kept. Your app records each step’s reads; Kam keeps those versions while retained.

Work gets redone.

Agents can’t tell what changed.They redo it all; a crash starts over.

  • Full rebuilds
  • Repeat calls
  • Restarted jobs

Only what a change touches reruns. Kam’s own work resumes after a crash. Resuming whole agent tasks is next.

Three ways to share results between agents

  1. 1

    Re-read every source, every turn

    Easy to start. Breaks quietly.

    • Exact version on request: no
    • A fix keeps the old version: no
    • A retry counts once: no
    • Only affected results update: no
    • Access checked on every read: you build it
    • Built for AgentCore: no

  2. 2

    Your database plus custom code

    Works, if you build it all.

    • Exact version on request: you build it
    • A fix keeps the old version: you build it
    • A retry counts once: you build it
    • Only affected results update: you build it
    • Access checked on every read: you build it
    • Built for AgentCore: you build it

  3. 3

    Kam

    Built in. In code today, not yet hosted.

    • Exact version on request: built in
    • A fix keeps the old version: built in
    • A retry counts once: built in
    • Only affected results update: built in
    • Access checked on every read: built in
    • Built for AgentCore: built in

means you build and maintain it yourself. Categories, not named products.

Start from your role
Implemented in source

AgentCore developer

Stop rebuilding State plumbing around your agents.

Explore history →Implemented in source

Multi-agent team

One versioned record any agent can read, with access checked.

Explore access →Implemented in source

AI infrastructure engineer

An old worker can’t overwrite a newer result.

Explore recovery →Planned · not available

Enterprise architect

Know exactly what changed, and what each step read.

Explore context →Planned · not available

Long-running workflow builder

Keep the work when a worker stops.

Explore memory →Planned · not available

AI platform team

Reuse finished work instead of computing it again.

Explore efficiency →
Start with Core. Grow into managed workflows.

  1. 01

    Kam Core

    Implemented in source · prelaunch

    Explore this layer

    Record, retry safely, correct, and read exact History and bounded State. SDK distribution and hosted access are not generally available.

  2. 02

    Managed State & Context

    In development · not hosted

    Explore this layer

    Prepare versioned evidence for the next agent, track dependencies, and reconcile stale results with explicit authority.

  3. 03

    Managed Workflows

    Product direction · not available

    Explore this layer

    Configure approved parent-child jobs, execution windows, adapters, and recovery rules. Kam determines eligible work from accepted State; execution can remain on AWS or customer infrastructure.

Follow a result from start to correction.

Try a software release or a fictional NFL game. Inspect versions, dependencies, access, and recovery yourself.

Explore the NFL exampleExplore a release

For AgentCore builders

Use exact State and History contracts instead of writing your own version tracking, retry handling, and evidence lookup.

See the code

Prelaunch, with AWS Marketplace planned

State and History are implemented in source. Hosted access and an AWS Marketplace listing are not available yet.

Review the evidence
Built to add on to AWS

Amazon Bedrock AgentCore

AWS runs the agent. Kam adds versioned State and exact work History beside it.

Your databases

Keep your records. Kam does not replace them.

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.