Show HN: CongaLine – Self-hosted isolated AI agent fleet (OpenClaw, Hermes)

4 points

2 months ago

We built CongaLine to solve a specific problem: give teams and individuals self-hosted AI assistants without collapsing everyone into a shared instance. Each agent runs in its own Docker container with isolated networks, secrets, and config. Security is the primary design constraint.

Three deployment targets via a single Go CLI (conga):

  - Local: Docker Desktop, no cloud required                                                                                                                   
  - Remote: any SSH-accessible Linux host (VPS, Raspberry Pi, bare metal)
  - AWS: zero-ingress via SSM, no inbound ports

Runtime is decoupled from deployment. OpenClaw (Node.js) and Hermes Agent (Python) ship out of the box - same infrastructure, your choice of agent. Multi-runtime is fully wired on local today; remote and AWS default to OpenClaw. More coming soon.

Channels are optional. Every agent gets a web UI via SSH/SSM tunnel. Slack and Telegram are both supported, each with a dedicated router container that fans events to per-agent containers. Two agent types: user agents (DM-based) and team agents (channel-based).

The name is literal: spiny lobsters travel in single-file conga lines during seasonal migration, in physical contact, reducing drag and sharing protection. A procession of isolated-but-coordinated agents.

Apache 2.0. Repo has Terraform for AWS, a Go CLI, and a Node.js Slack router.

Would love feedback on the provider abstraction. We're planning Kubernetes and GCP next.