FAME Architecture
A reference architecture — **F**unctions-as-a-Service-based **A**rchitecture for orchestrating **M**CP-**e**nabled agentic workflows — that decomposes complex agent reasoning into discrete, composable serverless entities (planners, actors, evaluators) orchestrated as step functions, with intelligent routing of state: lightweight conversational state into low-latency key-value stores (DynamoDB / Redis), heavy durable artifacts + cached tool responses into S3-compatible object storage. Published as arXiv 2601.14735 in early 2026.
Definition
A reference architecture — **F**unctions-as-a-Service-based **A**rchitecture for orchestrating **M**CP-**e**nabled agentic workflows — that decomposes complex agent reasoning into discrete, composable serverless entities (planners, actors, evaluators) orchestrated as step functions, with intelligent routing of state: lightweight conversational state into low-latency key-value stores (DynamoDB / Redis), heavy durable artifacts + cached tool responses into S3-compatible object storage. Published as arXiv 2601.14735 in early 2026.
Stateless FaaS environments (AWS Lambda, Cloud Functions, Azure Functions) offer pay-per-invocation economics and infinite scale-out, but the stateless contract is fundamentally hostile to multi-turn agent reasoning. Holding all state in the function payload bloats invocations and breaks at moderate context sizes; holding state in any single fast store doesn't scale across the heterogeneous data types agents produce (text, tool outputs, multimodal artifacts). FAME's bi-tier routing — hot state in KV, cold state in S3 — captures the structural property: agents have a small bounded *working* state plus a large unbounded *artifact* state, and they don't get accessed the same way.
Recent developments
- FAME paper published with empirical results. 13x latency reduction, 88% input-token reduction (from caching), 66% cost reduction on a research-paper-summarization benchmark. Per arXiv 2601.14735 — Optimizing FaaS Platforms for MCP-enabled Agentic Workflows.
- Adopted as a reference pattern in the AWS Builder Center. AWS's "Building AI Agents from Zero to Hero" walks through a FAME-style deployment using Lambda + DynamoDB + S3. Per AWS Builder Center — Building AI Agents.
- Complements Durable Agent Runtimes. FAME and Kitaru are not competitors — FAME is the deployment topology (which functions run where, which state goes where), Kitaru is the replay/resume primitive layered inside or alongside it. Many production systems combine both. Per Pydantic — Runtime layer for Pydantic AI agents.
Connections 8
Outbound 6
scoped_to1is_a1depends_on2complements1Inbound 2
is_a1constrained_by1