AgentCore Deployment

Deploy the buyer agent to Amazon Bedrock AgentCore as a managed runtime. The buyer handles campaign planning and budget allocation — seller interactions (inventory, pricing, deals) are handled by the seller runtime separately.

---

Prerequisites

• AWS CLI configured with credentials (aws configure or --profile)
• Python 3.12+ with pip install bedrock-agentcore
• No Docker required — CodeBuild builds ARM64 containers in the cloud

---

Quick Start

bash infra/aws/agentcore/deploy.sh \
  --mode http \
  --name my-buyer-agent \
  --profile my-aws-profile \
  --test

---

Architecture

The buyer runtime wraps DealBookingFlow in a BedrockAgentCoreApp container:

┌─────────────────────────────────────────────┐
│           AgentCore Container                │
│                                              │
│  ┌────────────────────────────────────────┐ │
│  │  BedrockAgentCoreApp (port 8080)       │ │
│  │  http_main.py                          │ │
│  │                                        │ │
│  │  ┌─────────┐    ┌──────────────────┐  │ │
│  │  │  crew   │    │     chat         │  │ │
│  │  │  mode   │    │     mode         │  │ │
│  │  └────┬────┘    └────────┬─────────┘  │ │
│  │       │                  │            │ │
│  │       ▼                  ▼            │ │
│  │  DealBookingFlow    ChatInterface     │ │
│  │       │                               │ │
│  │       ▼                               │ │
│  │  PortfolioCrew (Bedrock LLM)          │ │
│  │       │                               │ │
│  │       ▼                               │ │
│  │  Channel Specialists                  │ │
│  │  (CTV, Display, Mobile, Performance)  │ │
│  └────────────────────────────────────────┘ │
└─────────────────────────────────────────────┘

Crew Mode (Default)

Runs DealBookingFlow with the inner PortfolioCrew using Bedrock Converse. The crew allocates budget across channels based on the campaign brief.

curl -X POST http://localhost:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Plan a $500K Q4 automotive campaign across CTV and digital video"}'

Returns structured JSON with budget allocations, audience coverage, and approval status.

Chat Mode

Falls back to the existing ChatInterface keyword router.

---

Deploy Script

bash infra/aws/agentcore/deploy.sh [OPTIONS]

Options:
  --mode http           Runtime mode (HTTP only for buyer)
  --name NAME           Agent name
  --profile PROFILE     AWS CLI profile
  --region REGION       AWS region (default: us-west-2)
  --test                Run integration tests after deploy

Environment Variables

Variable	Default	Description
ROUTING_MODE	crew	Default routing mode
DEFAULT_LLM_MODEL	bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0	Bedrock model for PortfolioCrew
STORAGE_TYPE	sqlite	Storage backend
DATABASE_URL	sqlite:///:memory:	Database connection string

---

Campaign Brief Parsing

The buyer accepts natural language prompts and extracts structured campaign parameters:

Parameter	Extraction	Example
Budget	$500K → 500000, $2M → 2000000	"Plan a $500K campaign"
Dates	Q4 → Oct-Dec 2026	"Q4 automotive campaign"
Audience	targeting ... clause	"targeting adults 25-54"

The PortfolioCrew inside DealBookingFlow does the actual planning intelligence — budget allocation, channel research, audience coverage estimation.

---

Testing

# Unit tests (52 tests)
pytest tests/unit/agentcore/ -v

# Integration tests (3 tests, requires deployed runtime)
pytest tests/integration/agentcore/test_runtime.py \
  --profile genai --agent-name my-buyer-agent -v

---

Bedrock Converse Patch

Same patch as the seller — patches/crewai_bedrock_fix.py fixes orphaned toolUse/toolResult blocks and tool argument extraction in CrewAI's Bedrock provider.