Integrating SCL into AI Pipelines
SCL is the deterministic, cryptographic identity protocol for AI run declarations.
Architecture Placement
SCL is applied to the declared execution before runtime execution occurs. The integration point is at construction of the AI run declaration, prior to model invocation. Identity is derived from the declaration, not from outputs.
Pipeline Flow
A typical integration:
- Construct AI run declaration as SCL document bytes
- Validate, canonicalize, and hash using the reference engine
- Execute the model using the declared inputs
- Associate execution results with the precomputed SHA-256 identity
The identity remains stable and independent of execution outcome. For code-level implementation details, see using SCL in code.
Integration Targets
SCL integrates into:
- LangChain pipelines
- LlamaIndex workflows
- Custom inference servers
- Training loops
- Batch processing systems
In all cases, SCL operates on the declaration prior to execution.
Identity Binding
The SHA-256 identity binds execution results to a deterministic declaration identity. Two systems referring to the same declared execution produce the same identity.
Logging and Audit
SCL identity is recorded alongside execution metadata:
- SHA-256 identity stored in logs
- Identity associated with model inputs and configurations
- Identity referenced in audit records
Logs reference identity but do not define it.
Deterministic Guarantees
For identical input bytes, all compliant implementations MUST produce identical canonical JSON bytes and identical SHA-256 identity. See the SCL:V1 specification for complete rules.