Hybro Hub is a Python daemon that connects local AI agents on your machine to the hybro.ai portal, so local and cloud agents appear side by side in one web interface. It targets the trade-off between cloud platforms, which are capable but send data to third-party servers, and local runtimes, which keep data private but are isolated with no shared UI.
The hub runs as a background daemon linked to hybro.ai with an API key, and it discovers local agents and syncs them to the portal. In hybro.ai a local agent shows next to cloud agents and can be added to a room. Messages to local agents route through the relay to your hub, are processed on your machine, and only the response travels back, with a badge in the UI marking whether each message was handled locally or in the cloud.
The hub initiates all connections outbound over HTTPS, so it needs no inbound ports and works behind NAT. Local and cloud agents communicate over the Agent-to-Agent (A2A) protocol, and any A2A-compatible agent works. If the hub goes offline, cloud agents keep working while local agents show as offline and messages queue until it reconnects.
Local agents can be launched from bundled adapters for Ollama, OpenClaw, n8n, and Hermes, each with its own options, or any A2A agent can be auto-discovered on the machine by probing for an agent card, or registered manually in the config file. Configuration lives in a YAML file with environment-variable references, and a disk-backed publish queue retries events that fail to deliver.
Features
- Unified portal: local and cloud agents appear together in the hybro.ai web interface, chosen per conversation
- Local processing: messages to local agents are handled on your machine, and the cloud relay sees routing metadata rather than content
- Outbound-only connection: the hub initiates all traffic, needing no inbound ports and working behind NAT
- A2A protocol: local and cloud agents interoperate over the Agent-to-Agent protocol, and any A2A-compatible agent works
- Bundled adapters: launch Ollama, OpenClaw, n8n, or Hermes agents with adapter-specific options
- Agent discovery: auto-discovers local A2A agents via agent-card probing, with manual registration and LAN agents supported
- Sensitivity detection: scans outbound messages for PII, custom keywords, and regex patterns, currently logging detections only
- Graceful degradation: cloud agents keep working when the hub is offline, with local messages queued until reconnection
- Offline resilience: a disk-backed publish queue with size, TTL, and per-event retry limits
- Privacy indicators: each message in the portal is badged local or cloud based on where it was processed
