Quickref card#

One-page memgar reference.

Install#

pip install memgar                                   # core
pip install "memgar[feed,observability,gateway]"     # production
pip install -e ".[dev,adversarial]"                  # contributors

Analyze#

from memgar import Analyzer, MemoryEntry, Decision

a = Analyzer(use_llm=False)
result = a.analyze(MemoryEntry(content="..."))

result.decision         # Decision.ALLOW / QUARANTINE / BLOCK
result.risk_score       # 0-100
result.layers_used      # ['pattern_matching', 'transformer_ml', ...]
result.explanation      # human-readable
result.threats          # list[ThreatMatch]

fail-close#

a = Analyzer(fail_close=True)
# or: export MEMGAR_FAIL_CLOSE=true

Escalates ALLOW → QUARANTINE when any ML layer or feed is degraded.

Trust#

a.register_source_trust("openai-api",      0.95)
a.register_source_trust("user-form",       0.40)
a.register_source_trust("untrusted-wiki",  0.10)

Health#

a.health_check()
# {'patterns': {...}, 'layer1_5_semantic_guard': {...},
#  'layer2_ml_transformer': {...}, 'threat_feed': {...}}

Calibration (offline)#

# Strict gate
python scripts/calibrate_fpfn.py \
    --corpus ml/data/calibration_corpus.json \
    --output ml/artifacts/fpfn_calibration.json --no-llm
python scripts/check_calibration_gate.py

# Expanded gate
python scripts/calibrate_fpfn.py \
    --corpus ml/data/calibration_corpus.json \
    --corpus ml/data/mined_hard_subset.json \
    --corpus ml/data/augmented_memory_context.json \
    --output ml/artifacts/fpfn_calibration_expanded.json --no-llm
python scripts/check_expanded_gate.py

Environment#

Var	Default	Effect
`MEMGAR_FEED_ENABLED`	true	Pull signed threat feed
`MEMGAR_FEED_MAX_AGE_DAYS`	7	Cache staleness
`MEMGAR_OBSERVABILITY_ENABLED`	false	Prometheus + drift
`MEMGAR_OBSERVABILITY_PORT`	9090	Scrape port
`MEMGAR_FAIL_CLOSE`	false	Escalate on degraded
`MEMGAR_TRANSFORMER_THRESHOLD`	0.92	ML signal cutoff
`MEMGAR_CACHE_DIR`	`~/.cache/memgar`	Pattern + feed cache

CLI#

memgar feed sync        # force pull threat feed
memgar feed status      # last outcome + version
memgar feed verify      # signature check
memgar gateway run      # FastAPI gateway mode

Decisions#

`Decision`	When	Action
`ALLOW`	risk < 40	use as-is
`QUARANTINE`	40 ≤ risk < 80	audit / human review
`BLOCK`	risk ≥ 80 or CRITICAL match	reject

Pattern ID prefixes#

Prefix	Category
`INJ-`	prompt injection / override
`EXFIL-`	exfiltration / leak
`EXEC-`	execution / malware / cyber
`MANIP-`	manipulation / disinformation
`CRED-`	credential / token leak
`PRIV-`	privilege escalation
`FIN-`	financial fraud / wire transfer
`EVAS-`	encoding / obfuscation evasion
`BHV-`	sleeper / conditional behaviour

See the full catalog for every ID.