SCOPED ANALYSIS TASKS
What the model is allowed to do
Each pass is tightly scoped and its output must survive structural checks, binary constraints, static-versus-runtime comparison, and confidence gates. Fluent output that conflicts with the binary is rejected or downgraded.
›Candidate function classification
›Call-graph & cross-function reasoning
›Compiler-pattern interpretation
›Candidate type & role inference
›Ambiguity & contradiction detection
›Competing-hypothesis generation
›Adversarial critique of earlier interpretations
›Technical explanation & readability
AI proposes. Binary evidence disposes.
No model opinion enters the output until the disassembler, the IR, and the constraints agree.
USAGE POLICIES & SEPARATE CREDITS
Pay-as-you-go AI at provider list rate
External AI usage is billed separately with zero platform markup. You pay the provider's official rate for the models you choose to run.
OpenAI
Authorized API models · pay-as-you-go
Anthropic
Authorized API models · pay-as-you-go
Access routed through authorized OpenAI and Anthropic API models.
Model selection matched to analytical role, context size, latency, and cost.
AI credits are separate from the one-time software license.
Zero markup on the provider's published API list rate with pass-through billing.
Usage records show provider/model, metered units, rate basis, and charge.
Provider price changes flow through at the current official list rate.