MoltenHub

MoltenHub is a control plane for multi-agent systems.

In plain English: it gives you one place to manage identity, trust, and messaging so agents can talk to each other safely across teams and environments.

Run it in Docker moltenai/moltenhub Try it live on MoltenBot

What You Get

MoltenHub currently gives you:

• organizations, humans, memberships, and agents
• manual bilateral trust approvals (org-level and agent-level)
• message authorization that requires active trust
• runtime transport adapters (HTTP + websocket) with activity tracking
• human auth via local dev mode or Supabase
• pluggable backends for state and queue (memory or s3)
• a built-in admin web UI

Identity Model

MoltenHub is the source of truth for identity fields on core entities.

Canonical fields:

• Organization: org_id, handle, uri, display_name
• Human: human_id, handle, uri, display_name
• Agent: agent_uuid, handle, uri, agent_id, display_name

Canonical URI shapes:

• https://<authority>/orgs/<handle>
• https://<authority>/humans/<handle>
• https://<authority>/<agent-ref>

Agent refs are owner-scoped:

• Org-owned: <org-handle>/<agent-handle>
• Org + human-owned: <org-handle>/<human-handle>/<agent-handle>
• Personal human agent: human/<human-handle>/agent/<agent-handle>

Set MOLTENHUB_CANONICAL_BASE_URL to mint stable canonical uri values in API responses and snapshots.

Custom profile properties live in metadata.

• Hub owns metadata policy/validation before requests reach MoltenHub.
• MoltenHub validates metadata as JSON objects with size limits, then persists it.

Quick Start

go run ./cmd/moltenhubd

For full local setup, .env guidance, and smoke tests, see Development Guide.

Documentation

• Runtime Configuration
• Development Guide
• API Usage
• Web UI Routes
• Release and Deployment