Book a demo

Tell us about your use case. We will walk through composing a workflow, deploying to AWS Bedrock AgentCore, and invoking it from your product.

Deploy

How Floprints deploys WorkflowPlan agents to AWS Bedrock AgentCore — one harness per agent, versioned reconciliation, and ready polling.

What deploy does

When you deploy a version, Floprints:

  1. Creates or reconciles one Bedrock AgentCore harness per plan agent
  2. Wires each harness with its model, system prompt, and resolved tools
  3. Polls each harness to READY, then polls its runtime to READY
  4. Marks the version deployed and sets the blueprint’s deployedVersionId

Unchanged agents on redeploy reuse existing ready harnesses — they are not recreated unnecessarily.

Orchestrator harness

The orchestrator agent is deployed with a distinct tool set and a protocol-aware system prompt (wrapping the composer-authored role prompt) so the model emits tool calls instead of narrating them.

Orchestrator inline tools:

  • invoke_agent — call a specialist agent
  • parallel — fan out independent agents
  • ask_user_question — pause for human input
  • workflow_complete — finish successfully
  • fail — terminate with error

Prerequisites

Before deploy succeeds, ensure in Settings → Providers:

  • AWS region configured
  • Access keys stored (envelope-encrypted)
  • Execution role ARN set
  • Runtime container URI configured
  • Gateway ARN set (for gateway tools)

Group-scoped provider access must allow the deploying user’s groups to use the workflow’s provider (admins bypass).

Deploy status

Watch the version status in the workflow detail page:

  • deploying — harness reconciliation in progress
  • deployed — ready to run
  • failed — check audit log and gateway errors for details

Cost caps

Operators can only tighten the iteration/timeout envelope via environment variables (AGENT_MAX_ITERATIONS, AGENT_TIMEOUT_SECONDS, etc.). The console validates against the envelope and cannot silently raise limits above what the deployment allows.