Create robust binary AI runtimes
Paramus AI Service Specification 1.1
This specification defines a universal, lightweight interface for AI model deployment, monitoring, and interoperability. Use any language to implement, Python, C++ …
We expect 25+ AI models running concurrent
This specification standardizes REST-based endpoints for inference, health, metrics, and citation, enabling a reproducible and testable workflow for AI in science. Models using this design can be deployed as local executables, integrated cloud microservices, or registered modules inside the Paramus Marketplace.
The specification embodies Paramus principles: simplicity, observability, and interoperability. Each service provides self-describing capabilities, versioned responses, and full attribution to scientific and software sources.
It ensures that every Paramus model behaves predictably across all environments: stable system runtimes that cannot block each other.
Example
| Endpoint | Purpose | Example Response |
|---|---|---|
/health | Service status check | { "status": "ok", "model": "TransPolymer" } |
/capabilities | Declares supported tasks | { "capabilities": ["embedding","regression"] } |
/predict | Executes inference | { "success": true, "results": { "embedding": [...] } } |
/metrics | Reports uptime and latency | { "uptime_s": 3600, "avg_latency_ms": 45.2 } |
/citation | Provides research attribution | "Xu et al., npj Comput. Mater. 9, 64 (2023)" |