Enterprise Memory Infrastructure · Now in Research Preview

Enterprise Memory
That Knows Why

Persistent, traceable reasoning for AI systems in finance, legal, and compliance. Multi-hop retrieval with full decision provenance.

98.7%
Recall@5
HotpotQA
95.3%
Recall@5
2WikiMultiHop
81.5%
Recall@5
MuSiQue

Multi-hop retrieval benchmarks · Research evaluation results

Retrieval without memory is not intelligence

Standard RAG systems treat every query as a fresh start. They retrieve, answer, and forget. Enterprise decisions require more.

No Decision Traceability

When an AI system makes a recommendation, there is no way to trace which documents informed it, whether any were superseded, or why earlier evidence was discarded.

Multi-Hop Retrieval Fails

Complex questions require reasoning across multiple documents — "What regulation superseded the rule that this contract references?" Standard vector search collapses on these.

No Authority Awareness

In legal and regulatory documents, what matters is the controlling document — the one that is currently active. Semantic similarity retrieves the wrong answer confidently.

Architecture

A memory system built like the mind

Inspired by how divergent thinking works — capturing broadly, consolidating deliberately, retrieving with awareness of authority and context.

01

Capture Broadly

Every document is ingested without premature filtering. Content is embedded, described in context, and tagged — preserving raw information before structure is forced on it.

// Four encoding depths
L0 Raw passthrough
L1 Keywords + context
L2 Entities + relations
L3 Domain crystallization
02

Consolidate Intelligently

Background workers extract knowledge into a graph — deduplicating entities, discovering cross-domain bridges, and detecting contradictions between episodic memories.

// Background consolidation
merge deduplicate
bridge_scan cross-domain
decay age signals
schema_refine evolve model
03

Retrieve with Authority

Queries trigger two-phase retrieval: vector search fused with graph traversal via Personalized PageRank — surfacing controlling documents, not just similar ones.

// Hybrid retrieval
vector_ann similarity
graph_ppr graph walk
rrf_fusion rank merge
car_frontier authority
Input
Documents & Queries
Storage
Episodic Memory
PostgreSQL + pgvector
Knowledge Graph
Neo4j + GDS
Retrieval
Vector Search
HNSW ANN
Graph Traversal
Personalized PageRank
Output
Traced Answer with
Provenance + Authority
Capabilities

Built for the questions that matter most

Every feature is designed for high-stakes retrieval where wrong answers have real consequences.

Decision Traceability

Every retrieval result carries full provenance — which memories informed it, when they were captured, and whether any contradictions were detected along the way.

Multi-hop Reasoning

Answers complex questions that require chaining across multiple documents. Graph-based traversal discovers paths that pure embedding search cannot see.

Controlling Authority Retrieval

In authority-governed corpora (legal, regulatory), retrieves the active controlling document — not the most similar one. Knows when regulations are superseded.

Interest-Aware Context

The system learns domain interests over time. Retrieval results are weighted by relevance to tracked expert domains, giving analysts the context they actually need.

Flexible LLM Routing

Route different tasks to different models. Run fully local with Ollama at zero cost, or use cloud models for critical extraction. Cost profiles from $0 to $80/mo.

Contradiction Detection

The knowledge graph maintains CONTRADICTS edges. When a retrieved answer is in tension with stored knowledge, it surfaces the conflict — not just the most recent answer.

Markets

Where authority matters

Designed for domains where the wrong document — or the outdated one — has real consequences.

Finance

SEC filings, regulatory compliance, FINRA rules

  • Find the controlling regulatory event for any securities action
  • Trace compliance decisions to source filings with timestamps
  • Multi-hop queries across interconnected regulatory frameworks
Benchmark: FinSuperQA · SEC Authority Retrieval

Legal

Case law, precedent tracking, statutory authority

  • Identify superseding case law and active precedent chains
  • Answer multi-hop questions across statutes and case law
  • Full citation chain with contradiction detection
Benchmark: SCOTUS Precedent Retrieval

Security & Vulnerability

CVE tracking, patch releases, GHSA advisories

  • Track patch release timelines and dependency chains
  • Resolve the active advisory for a CVE across superseding fixes
  • Cross-domain correlation between vulnerability and compliance docs
Benchmark: GHSA Vulnerability Authority

Pharma & Life Sciences

FDA drug labels, clinical trial data, recalls

  • Retrieve active drug label versions, not superseded ones
  • Track recall cascades across drug families and interactions
  • Multi-hop reasoning across clinical and regulatory docs
Benchmark: FDA Label Supersession Retrieval
Research Foundation

Check some of our early public research

Early-stage public research — not peer-reviewed, but grounded in empirical benchmarks and real-world evaluation.

CAR Latest

Controlling Authority Retrieval

For authority-governed corpora, the retrieval target is the active frontier of the authority closure — not the highest-scoring document. Benchmarked across FinSuperQA, SCOTUS, GHSA, and FDA.

FinSuperQA
Finance · SEC
SCOTUS
Legal · Precedent
NoLLMRAG

Question-Space Dominance

Original question embeddings outperform bridge passages for multi-hop retrieval. Bridge passages contribute via entity identity, not embedding similarity — a counterintuitive finding with major efficiency implications.

98.7%
HotpotQA
95.3%
2WikiMH
81.5%
MuSiQue
BridgeRAG

Multi-hop Bridge Discovery

Systematic study of bridge passage retrieval strategies for complex multi-hop questions. Establishes baseline methods and failure modes that motivated Question-Space Dominance.

MDMA

Monotropic-Divergent Architecture

The production system combining all findings: neurodiversity-inspired memory architecture with episodic/semantic duality, interest tracking, and hybrid convergent/divergent retrieval.

Evo-Memory Evaluation Framework

Quality measured across 4 orthogonal axes

A
Answer Accuracy
F1 ≥ 0.70 factual
F1 ≥ 0.60 multi-hop
R
Retrieval Recall
Recall@5 ≥ 0.70
cross-domain ≥ 0.60
E
Step Efficiency
Overhead ≤ 3.0x
vs. naive baseline
S
Sequence Robustness
Std dev ≤ 0.10
across 5 orderings
Early Access

Be among the first

PortMem is in research preview. We are working with a small number of enterprise partners in finance and legal to validate the system in production environments.

No spam. We will reach out personally to discuss your use case.

4
Research papers
v0.1
Production system
4
Enterprise benchmarks