OpenHands - Hardened Self-Hosted AI Software Engineering Agent

This is a repackaged software product wherein additional charges apply for hardening, security configuration, and support.

WHAT IS OPENHANDS

OpenHands (formerly OpenDevin) is an open-source autonomous AI software engineering agent. The agent runs in a sandbox container with full read-write access to a workspace, executes shell commands, edits files, and drives a headless browser - controlled from a web IDE that streams live agent events over WebSocket. It works with any LLM provider through a configurable backend: Anthropic Claude, OpenAI GPT, AWS Bedrock, Azure OpenAI, Google Vertex AI, Cohere, Mistral, Groq, DeepSeek, Ollama, vLLM, LiteLLM, and more. Operators choose the provider in the web UI's Settings menu and paste their API key - no provider keys are baked into the AMI. The agent's CodeAct loop combines reasoning, tool use, and shell execution to perform real software engineering tasks: implementing features, debugging, refactoring, running tests, browsing documentation, and more. MIT license, no vendor lock-in.

WHAT THIS AMI ADDS

Security hardening:

• HTTP basic auth (admin / EC2 Instance ID) gates the entire app via Nginx - without this layer the OpenHands web UI is openly callable
• OpenHands container bound to 127.0.0.1:3000 - reachable only through the Nginx reverse proxy with TLS
• Sandbox agent-server runtime image pre-pulled and pinned by SHA-256 digest - reproducible builds, first conversation does not block on a 3-5 GB Docker pull
• Both images (openhands main + agent-server runtime) pinned by sha256 in /etc/openhands/openhands.env
• Nginx reverse proxy with TLS, HTTP-to-HTTPS redirect, WebSocket support for live agent events, security headers, large-body and long-timeout tuning for agent tasks
• No provider API keys baked in - operator configures their LLM provider through the OpenHands UI Settings after first login
• UFW firewall pre-configured - only TCP 22, 80, 443 are exposed
• fail2ban, AppArmor
• CVE scan - every image is scanned for vulnerabilities before release

OS hardening (CIS Level 1):

• CIS Ubuntu 24.04 LTS Level 1 benchmark applied via ansible-lockdown
• auditd, SSH hardening, kernel hardening, IMDSv2 enforced

Compliance artifacts:

• SBOM - CycloneDX 1.6 at /etc/lynxroute/sbom.json
• CIS Conformance Report at /etc/lynxroute/cis-report.html
• CIS Tailored Profile at /usr/share/doc/lynxroute/CIS_TAILORED_PROFILE.md

ARCHITECTURE NOTES

OpenHands spawns a sandbox container per conversation via /var/run/docker.sock mounted into the OpenHands container. This is required by the upstream design and gives the agent root-equivalent access to the host docker daemon. The AMI is therefore intended for single-tenant use - do not expose it to untrusted users. State, settings, and conversation history persist at /opt/openhands/state across container restarts and instance reboots. The agent's working files persist at /opt/openhands/workspace - operators can attach an EBS volume there for a larger or independently managed workspace.