pointlock-vision 0.1.5

Pointlock's Anthropic-backed visual verifier (downgrade-only, evidence-backed).
Documentation

pointlock-vision

Pointlock's Anthropic-backed visual verifier (downgrade-only, evidence-backed).

An optional visual verifier that speaks raw HTTP to the Anthropic Messages API. It may only downgrade a verdict's confidence against content-addressed evidence, never fabricate a pass.

Part of Pointlock — compile natural-language tasks into a typed, capability-bound, crash-safe flow IR. See the architecture spine for how this crate fits in.

License

Licensed under the Apache License 2.0.