Heavyweight suites
Fixed, elaborate metamodels and months of onboarding before the first useful view. Most organizations never get past the training phase.
Launching soon
Enterprise architecture your whole organization can contribute to. AI agents included.
seabove is a lightweight architecture repository built on a metamodel you define yourself. Application teams keep their part current in a lean client, AI agents fill the gaps, and your architects approve what enters. Start with a handful of types, not a training budget.
- Your metamodel, not a vendor's
- AI contributions under human review
- Hosted or self-deployed on PostgreSQL
- Billing Application
- Invoicing Capability
- Billing realizes Invoicing Realization
- Payment Gateway Application AI-provided · awaiting review
- PCI DSS Requirement AI-provided · approved
Why seabove
Between too heavy and too loose, there was nothing. Now there is.
Diagrams and spreadsheets
Quick to start, impossible to keep current. No shared structure a second team, a tool or an agent can build on.
Agents with nowhere to go
AI can research your landscape at scale, but its findings end up in chat windows, not in a repository your experts govern.
seabove gives you exactly as much structure as you define, and a front door for AI agents that your people control.
What you get at launch
Everything a growing EA practice needs on day one
A metamodel that fits your organization
Define element types, attribute types on base data types, and relationship types yourself. The metamodel is versioned and evolves forward-only, so a small start can grow toward standards such as ArchiMate without a migration project.
A repository where every entry conforms
Elements carry stable, unique names. Relationships connect only what the active metamodel allows. The structure you designed is the structure you get, from the first entry to the ten-thousandth.
A client people actually open
A lean, modern interface designed for application specialists, not just architects. Keeping your building blocks current takes minutes, not a certification.
A front door for AI agents
Agents read your metamodel and submit contributions over MCP, the Model Context Protocol, each with a declared confidence and rationale. Works with Claude Code and any other MCP client, out of the box.
Human review, built in
Every AI contribution stays marked as AI-provided until a specialist reviews it. High-confidence entries land directly; everything else waits in your review queue. You decide what becomes truth.
Runs where you run
One storage engine, PostgreSQL, from a developer's laptop to production. Use it hosted, or deploy it yourself and keep your architecture data inside your own perimeter.
Next on the roadmap after launch: automatic detection of logical errors in your model, and enrichment suggestions for existing entries.
How it works
From an empty repository to a governed model in four steps
-
1
Define your metamodel
Start with the few types you need today: Application, Capability, a relationship between them. Activate the version. Add more when your practice is ready, never before.
-
2
Model the landscape
Application teams add and maintain their building blocks in the client. Every entry conforms to the active metamodel, so the model stays coherent as more people contribute.
-
3
Let agents fill the gaps
Connect an AI agent over MCP. It reads the metamodel, researches your sources and proposes elements and relationships, each with confidence and rationale.
-
4
Approve what enters
Architects review the queue. Confirmed contributions lose their AI-provided mark and become part of the model; the rest are sent back or dropped. Humans stay in control.
Who it's for
Three kinds of contributors, one shared model
EA specialists
Shape the metamodel, curate the repository and approve proposals. Start small on day one and grow the model with your practice instead of ahead of it.
Application specialists
Own one or more building blocks and keep their information current in a client that feels like a modern web app, not an EA tool.
AI agents
Contribute what they find in documentation, tickets and code through a standard interface. Their entries are visible as AI-provided until a person confirms them.
Built to stay lean
Principles you can hold us to
- Headless first
- The reference client uses only the public interface. Anything you build, and any agent you connect, can too.
- Your metamodel, not ours
- No fixed ontology. A type's name is its stable identity; its label changes as freely as your vocabulary does.
- One storage engine
- PostgreSQL everywhere, so the setup on a laptop is the setup in production.
- Small by design
- A headless core and one lean client. Nothing to configure before the first useful view, and no module you install but never open.
Early access
Be first in line at launch
We are onboarding a small group of organizations before the public launch. Tell us about your practice and we will get in touch.