Back to Arena

Lyzr Cognis

by Lyzr AI

System Card

OrganizationLyzr AI
Released2024-10
Architecturehybrid / Context-Aware Write-Time Memory Consolidation
DetailsCognis is Lyzr's default memory layer for its enterprise agent platform, shipping with zero setup. Before storing any new memory, the system first retrieves similar existing memories and decides whether to add, update, delete, or skip — resolving contradictions at write time rather than at retrieval time. The open-source variant runs on Qdrant local + SQLite FTS5; the hosted version uses OpenSearch with Matryoshka embeddings and cross-encoder reranking. Backed by Accenture; $250M valuation.
Parameters—
Domainagent-memorylifelong-learningrag-retrieval
Open SourcePartial
WebsiteVisit
enterprise-agentsmemory-consolidationwrite-timebenchmark-sotaaccentureopen-source

Capability Profile

Benchmark Scores

2 of 14 benchmarks
Data Transparency:2 self-reported
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
1/2
LoCoMo
71.527pSelf-Reported
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
90.698pSelf-Reported
Multi-Hop QA
0/3
BABILong
no data
MultiHop-RAG
no data
HotpotQA
no data
Agent Task Memory
0/1
AgentBench-Mem
no data
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:Lyzr internal evaluation with GPT-4.1 — avg of 6 LongMemEval dimensionsLyzr internal evaluation — avg of 4 LoCoMo tasks vs Zep, OpenAI, Mem0