Recommendation trace
See the decision, not just the result.
Prism turns an unstructured request into an explainable recommendation your agent can inspect before it acts.
Analyzing prompt
Rust
Tool use
Code generation
Long context
Scoring eligible models
Each output includes the constraints applied, alternatives considered, and the evidence behind its score.
Architecture
A small, legible pipeline.
Prism keeps the path from a prompt to a decision explicit and composable.
- 01
Prompt
Extract intent, constraints, and task shape.
- 02
Capability analysis
Map requirements to a portable capability profile.
- 03
Policy engine
Apply your requirements for cost, latency, and providers.
- 04
Model scoring
Score eligible models against the complete profile.
- 05
Decision report
Return a choice with inspectable evidence.
Integrations
Policy is yours to define.
Profiles describe models without tying your decision system to a vendor.
Extend scoring with your own signals, budgets, and deployment constraints.
Use a structured decision report anywhere an agent needs to justify a choice.