Repository navigation

GLAIVE 0.3.0: benchmarks, privacy, audit trail, evidence search - #4

Merged

aliyaalias19 merged 21 commits into

mainfrom

feat/v0.3

Oct 3, 2026

Commits on Oct 3, 2026

fix(web): stop the server on the first Ctrl+C while a page is open
aliyaalias19
committed
9b0fbe7
docs: add roadmap for v0.3 to v1.0
aliyaalias19
committed
146cff1
feat(ingest): read NXLog and Winlogbeat JSON exports
aliyaalias19
committed
f19d1ae
feat(detection): tag alerts with ATT&CK tactics
aliyaalias19
committed
c0f6ee0
perf(detection): check each event only against rules for its log source
aliyaalias19
committed
64ba302
feat(ingest): extract .tar, .tar.gz and .tgz evidence archives
aliyaalias19
committed
1be9257
feat(bench): benchmark on public datasets with their own labels
aliyaalias19
committed
196b953
feat(eval): report confidence calibration against the answer key
aliyaalias19
committed
82155da
feat(privacy): pseudonymise case data before it reaches a cloud model
aliyaalias19
committed
14dd0ae
feat(obs): audit trail of every model call, tool call and agent run
aliyaalias19
committed
0a74381
fix(bench): remove read-only evidence copies after each case
aliyaalias19
committed
a7b5d32
feat(rag): hybrid evidence search over the graph (GraphRAG)
aliyaalias19
committed
4d91f31
feat(ask): answer questions from findings and evidence with checked citations
aliyaalias19
committed
0831427
feat(memory): opt-in past-case memory with indicator overlaps
aliyaalias19
committed
f05ee88
feat(bench): measure log poisoning (planted prompt injections)
aliyaalias19
committed
3f1ac0e
feat(security): detect prompt injection hidden by obfuscation
aliyaalias19
committed
98e8e39
feat(gate): hold AI claims that clear activity for analyst approval
aliyaalias19
committed
6f8dc63
feat(agents): Skeptic refutations of rule findings go to an analyst
aliyaalias19
committed
7592bb4
ci: run the public-data benchmarks on every push
aliyaalias19
committed
c538a49
docs: publish v0.3 accuracy results and document the new features
aliyaalias19
committed
0e028b8
chore(release): 0.3.0
aliyaalias19
committed
fcc5aab

Navigation Menu

Repository navigation

GLAIVE 0.3.0: benchmarks, privacy, audit trail, evidence search - #4

GLAIVE 0.3.0: benchmarks, privacy, audit trail, evidence search#4
aliyaalias19 merged 21 commits into
mainfrom
feat/v0.3

Commits on Oct 3, 2026

fix(web): stop the server on the first Ctrl+C while a page is open

docs: add roadmap for v0.3 to v1.0

feat(ingest): read NXLog and Winlogbeat JSON exports

feat(detection): tag alerts with ATT&CK tactics

perf(detection): check each event only against rules for its log source

feat(ingest): extract .tar, .tar.gz and .tgz evidence archives

feat(bench): benchmark on public datasets with their own labels

feat(eval): report confidence calibration against the answer key

feat(privacy): pseudonymise case data before it reaches a cloud model

feat(obs): audit trail of every model call, tool call and agent run

fix(bench): remove read-only evidence copies after each case

feat(rag): hybrid evidence search over the graph (GraphRAG)

feat(ask): answer questions from findings and evidence with checked citations

feat(memory): opt-in past-case memory with indicator overlaps

feat(bench): measure log poisoning (planted prompt injections)

feat(security): detect prompt injection hidden by obfuscation

feat(gate): hold AI claims that clear activity for analyst approval

feat(agents): Skeptic refutations of rule findings go to an analyst

ci: run the public-data benchmarks on every push

docs: publish v0.3 accuracy results and document the new features

chore(release): 0.3.0