01 / FOUNDATION
Data & semantics
Give agents durable facts, portable records, and shared meaning.
PLATFORM
A REFERENCE ARCHITECTURE FOR COMPOSABLE AI
An open agentic
platform is a stack,
not a suite.
Build agentic systems from open data foundations, model choice, interchangeable harnesses, and portable standards—without surrendering the seams.
REFERENCE STACK / 01COMPOSABLE BY DESIGN
04Open standards4 OPEN COMPONENTS
03Harnesses & brokers7 OPEN COMPONENTS
02Models & routing5 OPEN COMPONENTS
01Data & semantics5 OPEN COMPONENTS
YOUR POLICIESYOUR INFRASTRUCTUREYOUR CONTROL
DEFINITION / 00
Open components.
Explicit contracts.
Operational freedom.
An open agentic platform is an architecture in which data, models, execution, and interoperability remain independently understandable and replaceable. “Open” may describe source, weights, formats, or interfaces. A trustworthy architecture labels the difference instead of flattening it.
ARCHITECTURE / 01–04
Four layers.
No mandatory vendor.
Each layer answers a different question. Together they turn model capability into durable, governable work.
02 / INTELLIGENCE
Models & routing
Choose models by task, policy, economics, and deployment needs.
03 / EXECUTION
Harnesses & brokers
Turn intent into governed work with interchangeable runtimes.
04 / INTEROPERABILITY
Open standards
Make skills, context, profiles, and graphs portable across tools.
THE OPENNESS TEST / 05
Open is a property
of the whole system.
A pile of open-source parts can still produce a closed architecture. Test the relationships as carefully as the licenses.
Replaceable
Can one component be swapped without rebuilding the system?
Inspectable
Can a builder understand what runs and why?
Portable
Can identity, skills, context, and work move?
Bounded
Are authority and approval requirements explicit?
Grounded
Do agents share durable data and semantic meaning?
Auditable
Can people reconstruct decisions and outcomes?
A PRACTICAL PATH / 06
Build from the ground up.
Start with durable context. Add intelligence and execution only after control boundaries are clear.
- 1Ground the system
Choose open formats, a catalog, and a semantic layer that agents and people can share.
- 2Define the contracts
Express identity, skills, tools, workflows, policy, and approval points in portable forms.
- 3Compose the runtime
Select models, routers, brokers, and harnesses according to the work—not brand gravity.
- 4Observe and evolve
Retain evidence, evaluate outcomes, and replace components as requirements change.