Reasoning you can audit.
Ask complex questions. Follow every step. CLARITY breaks each answer into a verifiable chain — and shows you the receipts.
The reasoning pipeline
Five steps. Zero black boxes.
01
Parse
Normalize and tag every input.
02
Decompose
Split the problem into steps.
03
Draft
Write the draft with sources.
04
Verify
Score each claim, revise.
05
Deliver
Return the chain + confidence.
Models
Three sizes, one standard.
Compose-3
flagshipbenchmark · g-reason
92 / 100
latency 120ms
Evaluate-1
judgebenchmark · g-reason
87 / 100
latency 260ms
Micro-1B
edgebenchmark · g-reason
74 / 100
latency 34ms
The library
Modules that compose.
Retrieval
Grounded context from your own documents
Evaluation
Scores every answer against a rubric
Memory
Project memory that persists across runs
Tools
Call APIs, databases and spreadsheets
Guardrails
Hard rules the model cannot override
Traces
Step-by-step reasoning, fully auditable