# eCommerce store AI agents on Amazon Bedrock AgentCore

Companion artifacts for the FactualMinds post
**Build eCommerce Store AI Agents on Amazon Bedrock AgentCore (2026)**.

**Public URLs:** `https://www.factualminds.com/examples/architecture-blog-2026/ecommerce-agentcore-store-agents/{path}`

## Demo disclaimer

These files are a **sample architecture**, not a production commerce stack and **not** an anonymized client engagement.

- Order IDs (`ORD-1001`), SKUs (`SKU-TEE-BLU-M`), and dollar amounts are fake fixtures.
- Do **not** treat sample latencies or refund caps as measured eCommerce outcomes.
- First-party platform signals cited in the post (Gateway ~180→95 ms canary, AgentCore TCO worksheet) come from earlier published FactualMinds AgentCore posts — not from this demo.

## Versions (assumed)

Pin and verify in your account before deploy:

| Component | Assumed version / note |
| --------- | ---------------------- |
| Python | 3.12+ |
| `strands-agents` | ≥ 1.x (Strands Agents SDK) |
| `bedrock-agentcore` | AgentCore Python SDK matching Runtime HTTP contract |
| Model | `global.anthropic.claude-sonnet-4-5-20250929-v1:0` (swap per region) |
| AgentCore | Runtime + Gateway + Memory + Identity + Policy (mid-2026 platform) |

Local stubs run without AWS. Deployed Runtime needs IAM, Identity authorizer, and a Gateway target ARN.

## Layout

| Path | Purpose |
| ---- | ------- |
| [`architecture.drawio`](architecture.drawio) | Supervisor + 4 specialists + Gateway/Policy/Memory |
| [`routing-decision-matrix.md`](routing-decision-matrix.md) | Intent → agent → tools → Policy risk |
| [`supervisor_agent.py`](supervisor_agent.py) | Strands supervisor with route tools |
| [`specialists/`](specialists/) | Sales, orders, support, inventory stubs |
| [`gateway/commerce-openapi.yaml`](gateway/commerce-openapi.yaml) | OpenAPI Gateway target sketch (8 tools) |
| [`policy/refund-and-cancel.cedar`](policy/refund-and-cancel.cedar) | Sample Cedar write-path gates |
| [`monday-checklist.md`](monday-checklist.md) | Deploy / test gate list |

## Quick local dry-run

```bash
# From this folder — syntax-check only (no AWS calls)
python3 -m py_compile supervisor_agent.py specialists/*.py
```

Wire each specialist as its own AgentCore Runtime (or MCP target) and point the supervisor route tools at those ARNs. Attach `commerce-openapi.yaml` as a Gateway OpenAPI target and attach `refund-and-cancel.cedar` (or NL→Cedar equivalents) with Policy mode `LOG_ONLY` first, then `ENFORCE`.

## Related posts

- [AgentCore production guide](https://www.factualminds.com/blog/amazon-bedrock-agentcore-production/)
- [Gateway server-side tools](https://www.factualminds.com/blog/amazon-bedrock-agentcore-gateway-server-side-tool-execution-2026/)
- [Multi-agent supervisor pattern](https://www.factualminds.com/blog/aws-bedrock-multi-agent-supervisor-pattern/)