ElephantBroker: A Knowledge-Grounded Cognitive Runtime for Trustworthy AI Agents

Cristian Lupascu; Alexandru Lupascu

cs.AIMar 26, 2026

ElephantBroker: A Knowledge-Grounded Cognitive Runtime for Trustworthy AI Agents

Authors: Cristian Lupascu, Alexandru Lupascu

Organizations: Elephant Broker, 050141 Bucharest, Romania

Abstract

Large Language Model based agents increasingly operate in high stakes, multi turn settings where factual grounding is critical, yet their memory systems typically rely on flat key value stores or plain vector retrieval with no mechanism to track the provenance or trustworthiness of stored knowledge. We present ElephantBroker, an open source cognitive runtime that unifies a Neo4j knowledge graph with a Qdrant vector store through the Cognee SDK to provide durable, verifiable agent memory. The system implements a complete cognitive loop (store, retrieve, score, compose, protect, learn) comprising a hybrid five source retrieval pipeline, an eleven dimension competitive scoring engine for budget constrained context assembly, a four state evidence verification model, a five stage context lifecycle with goal aware assembly and continuous compaction, a six layer cheap first guard pipeline for safety enforcement, an AI firewall providing enforceable tool call interception and multi tier safety scanning, a nine stage consolidation engine that strengthens useful patterns while decaying noise, and a numeric authority model governing multi organization identity with hierarchical access control. Architectural validation through a comprehensive test suite of over 2,200 tests spanning unit, integration, and end to end levels confirms subsystem correctness. The modular design supports three deployment tiers, five profile presets with inheritance, multi gateway isolation, and a management dashboard for human oversight, enabling configurations from lightweight memory only agents to full cognitive runtimes with enterprise grade safety and auditability.

Figures & tables

Capability	Ours	Zep	Mem0	MemGPT	L.C.	QMD	S.M.
Knowledge graph	✓	✓	Opt.	—	—	—	—
Vector store	✓	✓	✓	—	—	✓	✓
Evidence tracking	✓	—	—	—	—	—	—
Goal-aware scoring	✓	—	—	—	—	—	—
Working set budget	✓	—	—	✓	—	—	—
Multi-agent actors	✓	—	Part.	—	—	—	Part.

Table 1: Comparison of agent memory systems across key architectural dimensions.

Panel of Figure 1 — Figure 1: Four-layer architecture of ElephantBroker. Thin TypeScript adapters in Layer A forward requests via HTTP to the Python runtime in Layer B, which delegates storage and search to the Cognee knowledge plane in Layer C, backed by the infrastructure services in Layer D.

Scope	Visibility	Decay Cadence	Promotion Target
GLOBAL	All gateways in org	Monthly	Terminal
ORGANIZATION	Matching org_id	Weekly	GLOBAL
TEAM	Matching team_id	Weekly	ORGANIZATION
ACTOR	Owner actor only	Daily	TEAM
SESSION	Current session	Hourly	ACTOR
TASK	Current task	Hourly	SESSION

Table 2: Eight-value scope system governing fact lifecycle.

Class	Typical Content	Decay	Promotable	Compaction
EPISODIC	Events, decisions	Yes	SEMANTIC (3+ sessions)	Summarizable
SEMANTIC	Durable knowledge	Slow	Terminal	Mergeable
PROCEDURAL	Workflows, SOPs	None	Terminal	Preserve
POLICY	Constraints, rules	None	Terminal	Always inject
WORKING_MEMORY	Transient context	Session-only	—	Drop at session end

Table 3: Five memory classes with lifecycle properties.

Entity Type	Schema Source	Indexed Fields	Vector Collection
Fact	Fact assertion	text	Fact_text
Actor	Actor reference	display name	Actor_display_name
Goal	Goal state	title, description	Goal_title
Procedure	Procedure def.	name, description	Procedure_name
Claim	Claim record	claim text	Claim_claim_text
Evidence	Evidence ref.	reference value	Evidence_ref_value

Table 4: Seven entity subclasses mapping schema models to graph–vector storage.

Panel of Figure 2 — Figure 2: Claim verification state machine with associated confidence multipliers $m_{v}$ . Each transition is driven by the type of evidence attached. Dashed arrows indicate explicit rejection, which is possible from any non-terminal state.

Parameter	Coding	Research	Manager	Worker	Personal
turn_relevance	1.5	0.8	0.7	1.3	1.0
session_goal_rel.	1.2	1.0	1.5	1.4	0.8
global_goal_rel.	0.3	0.8	1.0	0.3	0.4
recency	1.2	0.5	0.6	1.3	0.9
successful_use	0.8	0.6	0.5	0.7	0.9
confidence	0.3	0.8	0.5	0.4	0.3

Table 5: Scoring weights and configuration parameters across five profile presets.

Dimension	Procedures	Guards	Evidence
Nature	Prescriptive	Restrictive	Accountable
Timing	When activated	Every action	After action
Question	How to do it?	Is it safe?	Can you prove it?
Scope	Per-workflow	Per-session	Per-claim
Failure mode	Incomplete step	Blocked action	Unverified claim
Storage	Cognee graph	Redis + graph	Cognee graph

Table 6: Comparison of procedures, guards, and evidence across eight dimensions.

LLM Agent Memory

Large language model agent memory lets agents retain, organize, retrieve, and update information across interactions and long-running tasks.

See all 373 papers

Eywa: Provenance-Grounded Long-Term Memory for AI Agents
May 29, 2026Resham JoshiLLM Agent Memory Persistent Memory for Language Models
Eywa
Agent Zero Memory: Provenance-Aware Long-Term Memory for LLM Agents
Aug 30, 2026Ming Wu, Pengyuan ZhuLLM Agent Memory Long-Term Conversational Memory
Zero Labs
Agent-Sentry: Bounding LLM Agents via Execution Provenance
Mar 24, 2026Rohan Sequeira, Stavros Damianakis, Umar Iqbal +1Computer-Use Agent Safety AI Agent Security
University of Southern California, Los Angeles, California, USA · Washington University in St. Louis, St. Louis, Missouri, USA

May 29, 2026J/K move · Enter open · S save

Eywa: Provenance-Grounded Long-Term Memory for AI Agents

Resham Joshi

Eywa

LLM Agent Memory Persistent Memory for Language Models Long-Term Conversational Memory Agent Memory Management

Open paper AskPDF arXiv

AI agents that persist across sessions need memory they can retrieve, audit, update, and erase. Existing memory systems often collapse source evidence, extracted facts, retrieved context, and answer policy into one opaque prompt path, making failures difficult to diagnose: a wrong answer may come from missing evidence, unsupported extraction, stale state, retrieval loss, or answer-model behavior. We present Eywa, a provenance-grounded memory architecture built around evidence before belief. Eywa stores immutable source evidence before deriving canonical facts, validates extracted memories against typed signals and source support, and retrieves bounded memory context through a deterministic multi-route read path with zero LLM calls inside retrieval. Retrieved context is returned separately from answer instructions, allowing the same memory substrate to be evaluated across frontier, budget, and local answer models. Under a frozen, artifact-recorded retrieval configuration, Eywa reaches 90.19% judge accuracy on the LoCoMo C1-C4 split with Claude Sonnet 4.6 write and QA roles. On LongMemEval-S, it reaches 88.2% retrieval-sufficiency accuracy. On BEAM, a 700-question technical-memory stress benchmark, it reaches 81.45% mean nugget score and 85.29% pass@score >= 0.5. Full per-question artifacts, including questions, gold answers, model answers, retrieved context, and labels, are published at https://eywa.to/research.

Abstract

LLM Agent Memory

Explore similar work

Eywa: Provenance-Grounded Long-Term Memory for AI Agents

Agent Zero Memory: Provenance-Aware Long-Term Memory for LLM Agents

Agent-Sentry: Bounding LLM Agents via Execution Provenance