EXIA GHOST
0%

Cognitive Memory Architecture for Autonomous AI Agents

Remember. Reason. Resist hallucination.

Scroll to explore
COGNITIVE MEMORY — HALLUCINATION-RESISTANT — PROVEN BY BENCHMARKS — 

The brain
of autonomous AI

EXIA GHOST is not a memory add-on. It is a complete cognitive architecture — 8-stage pipeline, causal reasoning, anti-hallucination contract, ethical guardian. Built for agents that must remember, decide, and resist fabrication.

0 % HaluMem F1 (first independent evaluation)
0 % LoCoMo full pipeline, cats 1-4 (corrected)
0 % Update hallucination rate
0 $ External funding

First independent
evaluation

F1 71.99% — first independent evaluation of the HaluMem benchmark. Precision 92.90%, recall 58.76% (weighted 75.94%).

Judge: GPT-4o | 1 user, 65 sessions | Raw scores published in repo

71.52% on adversarial
questions (Category 5)

85.8% across all 5 categories; 84.16% full pipeline (cats 1-4, corrected); Category 5 adversarial: 319/446 = 71.52%.

10 conversations | Multi-hop, temporal, world knowledge, adversarial | Raw per-question data published

Independently evaluated,
self-funded

System
Funding
EXIA GHOST
71.99% HaluMem
84.16% LoCoMo
$0
MemOS
79.70% HaluMem
academic
Mem0
57.85%
66.9%
$24M
MemMachine
84.87%
$43.5M
Zep
<50%
58-75%
$2.3M

Critical Grade
cognitive pipeline

A complete cognitive memory architecture for autonomous AI agents. Five specialized memory types work in concert with causal emergence and recursive self-improvement — each cycle sharpens the next. An independent safety and ethics guardian operates outside the main pipeline, with veto authority over every output. The result: a cognitive pipeline that wraps the LLM, controls its behavior, and reduces hallucinations at the architectural level.

Patent pending (INPI France) | Proven across 6 LLM backends

EXIA GHOST is available for licensing, integration, and strategic partnership.