SYSTEM Cited by 2 sources

AgentFlo¶

What it is¶

AgentFlo is the agentic-commerce service by Salesflo — a platform for deploying always-on AI sales, support, and ordering agents across messaging channels (WhatsApp, SMS, RCS, web chat, voice). Its agents understand customer intent, connect to commerce systems, recommend products, create carts, and convert conversations into completed transactions. Per Salesflo, AgentFlo serves eCommerce merchants managing over $300 billion in annual transacted value across Shopify, WooCommerce, Magento, and SAP.

It is built entirely on AWS, organized around five pillars of production-grade AI agents: Velocity, Standardization, Scalability, Trust, and Reliability.

Architecture (as disclosed)¶

Messaging layer — AWS Fargate behind an ALB: authentication, image OCR, speech-to-text/text-to- speech (Opus OGG → MP3 with fuzzy product-name matching), pre-turn guardrails, prompt-injection detection.
Agent orchestration — AgentCore Runtime running the Strands Agents SDK; each session is an isolated microVM (concepts/micro-vm-isolation), stateful up to 8 hours.
Tool layer — AgentCore Gateway over MCP brokering calls to Lambda functions (Cart, Product, Knowledge Base) with IAM-based auth; OAuth/credentials held in the Gateway (patterns/central-proxy-choke-point).
Data layer — DynamoDB (sessions + carts), Aurora (orders), an Amazon Bedrock Knowledge Base on S3 for RAG.
Governance — AgentCore Policy (Cedar, deterministic tool authorization) with Bedrock Guardrails embeddable in Policy.
Observability/analytics — AgentCore Observability for logs/traces; Amazon Data Firehose → S3 for cost/revenue analytics.

Design choices worth noting¶

Recipe-based deployment — merchants pick a pre-built recipe (sales, restaurant ordering, clinic receptionist, support, B2B reorder, cart recovery) and go live on WhatsApp in minutes via a portal → GitHub Actions → AgentCore pipeline (recipe-based-agent-deployment).
Single domain-specialized agent over multi-agent for most interactions (single-agent-over-multi-agent), with clean handoff to a specialist when a conversation shifts domain (sales → support).
Model-driven architecture — tools as functions + system prompt, model orchestrates (model-driven-agent-architecture).

Trust & Reliability (Part 2, 2026-08-21)¶

Trust — safe action, not just safe responses: three-layer guardrails (three-layer-agent-guardrails) across pre-request (Fargate: prompt-injection detection + opt-outs + phone-number auth), tool execution (Policy Cedar rules + Gateway identity/order-locks with embedded Guardrails), and post-turn output screening; plus microVM isolation, IAM, VPC domain restrictions, and secrets in Secrets Manager (none in code). AgentCore Observability → CloudWatch gives per-turn traces, P50/P95 latency, and per-merchant/agent/conversation cost.
Reliability — data foundation that keeps agents grounded: two-table DynamoDB state with intent-gated cart loading (intent-gated-state-loading, concepts/agent-memory); Bedrock Knowledge Base grounding; semantic product discovery via Aurora embeddings / S3 Vectors; Data Firehose→S3 for cost-vs-revenue ROI.
Results (90-day control-group): +12% net revenue, +40% engagement, +15% conversion, +8% AOV, +20% reactivation.
Roadmap: real-time voice (two-pass domain-aware transcription over 90+ languages two-pass-domain-aware-transcription; decoupled STT / reasoning / TTS pipeline-component-decoupling; BidiAgent on AgentCore WebRTC + Kinesis Video Streams) and server-side tool execution (server-side-tool-execution, ~30% lower latency for short tool loops).

Caveats¶

Described in a two-part AWS Architecture Blog series: Part 1 covers Velocity, Standardization, Scalability; Part 2 covers Trust, Reliability, business results, and the roadmap. Both are now ingested.
Vendor/customer architecture narrative; no internal latency/throughput/cost numbers for AgentCore surfaces. Voice agents are in pilot; server-side tool execution is experimental; business-impact percentages are Salesflo-reported over an early 90-day window.

Seen in¶

sources/2026-08-20-aws-how-agentflo-built-ai-sales-agents-with-amazon-bedrock-agentcore — the Part 1 architecture post (Velocity / Standardization / Scalability).
sources/2026-08-21-aws-how-agentflo-built-ai-sales-agents-with-amazon-bedrock-agentcore-part-2 — Part 2 (Trust / Reliability + business results + voice / server-side tool-execution roadmap).