hermes-agent-aws: Private AI Agents on AWS

hermes-agent-aws deploys a private, always-on AI agent on AWS in one command — persistent memory, free-LLM failover, and a REST API for ~$17/month. Built by Shahriar Labs, it solves the "always-on agent" problem: you need an agent that's reachable, remembers context, and doesn't reset state between calls — without paying $100+/month for a managed agent platform.

Architecture

hermes-agent-aws provisions:

• ECS Fargate: Containerized agent runtime. Scales to zero when idle, spins up in ~30s on cold start. Stateless compute — state lives in DynamoDB.
• DynamoDB: Persistent memory store. Session memory (TTL-based), long-term memory (no TTL), and tool output cache (short TTL).
• API Gateway: REST endpoint for agent invocation. Supports sync (wait for response) and async (queue task, poll for result) modes.
• SQS: Task queue for async invocations and long-running agent jobs.

Infrastructure is Terraform-provisioned. One terraform apply deploys the full stack. Teardown with terraform destroy.

Free-LLM Failover

hermes-agent-aws integrates openrouter-free routing: primary calls go to your paid model (Claude, GPT-4), and fallback paths automatically route to free models (DeepSeek R1, Qwen 3) for non-critical tasks or when the primary is rate-limited. This keeps the agent operational at all times and reduces API costs by 40–60% for high-volume use cases.

When to Use hermes-agent-aws

Use it when you need: an agent that integrates deeply with your internal AWS infrastructure (RDS, S3, Lambda), persistent memory across sessions for a specific user or workflow, a private endpoint without sharing data with hosted agent providers, or always-on availability for business-critical agent tasks. For multi-tenant SaaS architectures see our post on multi-tenant SaaS on AWS ECS.

Frequently Asked Questions

What is hermes-agent-aws?

An AI skill that deploys a private always-on AWS agent (ECS + DynamoDB + API Gateway) with persistent memory for ~$17/month.

Why run private instead of hosted?

Data control, custom tool integrations, no usage limits, consistent availability.

What does it cost?

~$17/month baseline AWS infra. LLM API costs extra — free-model fallback reduces this significantly.

What AWS services are used?

ECS Fargate, DynamoDB, API Gateway, SQS, Secrets Manager, CloudWatch.