Engram — a trust layer for agent memory. Memories are claims with a lifecycle and a provenance graph; retrieval ranks by similarity x earned trust; quarantine cascades through graphLookup.
-
Updated
Aug 16, 2026 - Python
Engram — a trust layer for agent memory. Memories are claims with a lifecycle and a provenance graph; retrieval ranks by similarity x earned trust; quarantine cascades through graphLookup.
Sample-to-story malware analysis: CAPEv2/Sysmon report replay into provenance graphs and ATT&CK timelines, family attribution, and single-run Sigma/YARA synthesis with measured sibling recall and cross-family/benign FPR; standalone explainable EMBER gate. Lab-only, never executes samples.
🗺️ Interactive visual atlas dissecting APT lifecycles, causal provenance graphs, and detection telemetry.
Forensic timeline reconstruction from GUID-less artefacts: calibrated causal rules, confidence-graded incident stories and SHA-256 custody (Sysmon/EVTX, plaso, Volatility, auth.log, auditd). Lab-only DFIR.
Explainable provenance-graph attack reconstruction: time-respecting backtracking ablated on ATLAS with held-out host logs, a bpftrace probe run live in CI, Sysmon/auditd/ATLAS loaders, STIX/Neo4j export.
Evidence-first security knowledge graph: 13 security tools write reliability-graded claims into one graph, and fusing only independent engines ranks the true ATT&CK technique first more often and makes confident attribution right more often - measured with intervals on OTRF, APT29 and ATT&CK data
To associate your repository with the provenance-graph topic, visit your repo's landing page and select "manage topics."