PRODUCT_VISION.md

Vision

Build the operating layer that lets AI agents access real systems safely.

The world does not only need more agents. It needs a safer way for those agents to connect to:

• APIs
• email
• calendars
• CRMs
• commerce systems
• billing tools
• internal apps
• eventually payment systems

Product thesis

The market gap is not "talking AI." The gap is:

• safe execution
• secret abstraction
• policy
• approvals
• auditability
• local-first trust boundaries

Core user promise

The user should be able to say:

My agent can do useful work, but it cannot see or export my service secrets.

Who it is for

Community / builders

• self-hosters
• agent tinkerers
• AI workflow builders
• developers using local agents or scripted automations

Professional users

• agencies
• operators
• small businesses
• technical founders
• internal ops teams

Later expansion

• multi-site businesses
• compliance-minded teams
• managed service providers
• enterprise IT

User story

A user installs Harbor Node on a local device. They connect services through Harbor Ports. Their agent can now ask Harbor for approved actions.

Instead of:

• "give me the Stripe key"
• "send this arbitrary HTTP request"

The agent asks:

• "create invoice draft"
• "list today's open leads"
• "schedule follow-up message"
• "sync order status"

Harbor Guard decides whether to allow it.

Product principles

1. Secrets stay local whenever possible.
2. The local node is the real product.
3. The cloud is optional and additive.
4. Actions are named and governed.
5. Policy and audit are first-class.
6. Payment capability belongs in the architecture, not the early go-to-market surface.