makina prepares source-code security evidence before it is handed to an LLM. It is built for strong static analysis: rule-based scanning, semantic matching, taint tracing, structural checks, and call-graph context, with a continuously learning confidence model that retrains from human Review labels.
-- analyzes pasted code or dropped folders and turns detector output into deduplicated findings with line evidence, CWE metadata, and confidence scores.
-- shows the call-graph evidence behind findings, linking sources, functions, sinks, and reported issues in one inspectable workspace.
-- sends bounded scanner evidence through backend-managed LLM workflows and renders one structured MAKINA report block per finding.
-- exports the same structured report into a branded security report format suitable for sharing or archival.
-- is the human triage queue. Findings are marked TP/FP or closed here, and submitted decisions become training data.
-- keeps reviewed and closed cases available for later inspection, including source context, labels, and historical decisions.
-- shows the current learning state, accumulated labels, validation metrics, and retraining status.
