Research preview · Accepting design partners in finance and legal

The LLM retrieval layer for Financial and auditable regulated work.

We give LLMs superpowers.

The cheap LLM version with PortMem hits 99% accuracy on regulated questions, compared to 65% using it alone. If you prefer to use the latest agentic LLM that gets most answers right, PortMem makes querying it much cheaper. It is a win-win no matter which LLM you pick.

Speed and cost

Answering a question should not require ten LLM calls and thirty seconds.

The two dominant ways to "fix" RAG today are agentic loops (the model thinks, calls a tool, re-reads, repeats) and long-context stuffing (drop the whole corpus in the prompt and pray). Both pay a tax in tokens, dollars, and seconds, and neither solves currency or authority.

PortMem returns a verified passage in one retrieval round. The LLM, if you use one, only sees the right document.

10×

Faster than long-context.

Sub-second retrieval against your corpus, regardless of size. Stuffing 100k tokens into a frontier model is 20 to 40 seconds and grows with corpus.

10 to 100×

Cheaper per query.

A retrieval round is fractions of a cent. A long-context call on a frontier model is dollars per query. Multiply by daily volume.

1

One retrieval round.

Not five agent steps, not three rerank passes. Cost and latency are predictable, which is what compliance and procurement care about.

The solution

The retrieval is the answer.

Instead of asking an LLM to summarize what it found, PortMem returns the exact passage that should govern the answer, along with the trail of how it got there. The LLM, if you use one at all, becomes a presentation layer over a verified source.

01

Surface the source, not a summary.

The answer is a passage from a real document, with a citation. Nothing is generated on the critical path.

02

Currency and authority are filters, not vibes.

Superseded, recalled, and overruled documents are removed at retrieval time. The result is the one that is still in force.

03

Calibrated abstention.

When the evidence is weak or contradictory, the system says so. Better silence than a confident wrong answer in regulated work.

04

Multi-hop by construction.

For questions that need two or three connected documents, PortMem chains the retrievals and shows the bridge between them.

05

Per-query routing.

Different question types need different strategies. A router picks the right one without you having to label queries.

06

Drops onto what you already have.

Sits above your vector store. Vendor-neutral, model-agnostic, no rip-and-replace.

Where PortMem fits

Most stacks pick one axis. Regulated buyers need all of them.

PortMem sits above any vector store. Buyers add the regulated-domain ranking layer that long-context LLMs, agentic frameworks, and generic rerankers do not have.

← swipe to scroll →
Approach Currency Audit Multi-hop Model-agnostic Regulated Speed Cost
Long-context LLMs
Sonnet 1M, GPT-4
~ Slow $$$
Naive RAG
LangChain, LlamaIndex
~ Fast $
Agentic RAG
LangGraph, CrewAI, AutoGen
~ Slow $$
Compilation-stage knowledge
Pinecone Nexus, PageIndex
~ Med $$
Vertical legal AI
Harvey, Casetext, Hebbia
~ ~ legal only Med $$$
Enterprise search
Glean, Sana
~ Fast $$
Vector store + rerank
Vectara, Pinecone, Cohere
~ ~ Fast $$
PortMem
Retrieval for regulated markets
Fast $
Built and load-bearing ~ Partial or workaround ✗ Not built
The proof

Three regulated benchmarks. One pattern.

We built three test corpora where the right answer is not the most similar document. In every one, a workhorse frontier model working alone fails most of the time. PortMem clears it.

Finance · primary

Which version is in force?

100% PortMem
65% Workhorse frontier model alone

Across SEC filings, FASB updates, FINRA rule changes, and SEC no-action letters, the latest superseding document wins. PortMem picks it deterministically. A frontier LLM working from raw text gets it right 65% of the time.

Legal

Has this precedent been overruled?

99% PortMem
7% Workhorse frontier model alone

SCOTUS overrules are semantically distant from the cases they replace. PortMem finds them by authority structure, not by similarity.

Pharma

Has this drug been recalled?

99% PortMem
65% Workhorse frontier model alone

FDA's recall database overlaps with marketing-approval text. Standard retrieval pulls the approval; we surface the recall. Pharma demo →

Datasets, code, and the single-command evaluator are published with each paper. "Workhorse frontier model alone" baseline is Claude Haiku reading the same corpus directly without PortMem retrieval. Method, prompts, and full result tables are in the CAR paper (paper 4) under github.com/andremir/car-retrieval.

Markets

Finance first. Then legal, pharma, and security.

The pain is sharpest where a stale rule is an SEC enforcement action, a withdrawn no-action letter is a deficient supervisory procedure, or a missed supersession is a restated filing. We are starting with finance because the buyer has the budget, the audit clock is recurring, and the data (10-Ks, FASB ASUs, FINRA rule amendments, no-action letters) is structurally version-controlled and machine-readable.

Primary vertical

Finance

SEC filings, FASB Accounting Standards Updates, FINRA rule amendments, SEC no-action letters, internal compliance memos. Knowing which version is currently in force is the entire job, and the cost of being wrong is regulatory exposure measured in seven figures.

PortMem hits 100% accuracy on financial supersessions. A workhorse frontier model alone hits 65%. The 35-point gap is the gap between "tooling" and "audit-defensible."

Buyer roles
  • Chief Compliance Officer
  • Controller and SEC reporting
  • Internal audit and risk
  • Investor relations
Live workflows
  • 10-K and 10-Q supersession tracking
  • FASB ASU adoption status
  • No-action letter currency
  • FINRA rule amendment audit
Also supports

Legal

Case law, precedent tracking, statutory authority. One mis-cited brief is 80 to 200 hours of associate time and a malpractice exposure. PortMem catches overruled precedent that frontier LLMs miss 93% of the time.

Buyers: AmLaw partners, in-house GC, litigation support, legal AI platforms.

Pharma and life sciences

FDA labels, clinical trial protocols, drug recall notices. A recalled product retrieved as "approved" is a patient safety event.

Buyers: medical affairs, regulatory affairs, pharmacovigilance. Pharma demo →

Security and vulnerability

CVE entries, GHSA advisories, vendor patch notes. "Is this CVE patched?" is a controlling-authority question. Standard RAG gets it wrong 39% of the time.

Buyers: secops, vulnerability management, AppSec platforms.

Research foundation

Four arXiv papers. One company.

PortMem is the productization of four sole-authored research contributions. Each one fixes a specific failure mode that standard retrieval has in regulated content.

Paper 1 · Calibrated Fusion

PhaseGraph

Vector search and graph search produce scores on different scales. PhaseGraph maps them onto a common rank-based scale before fusing, which lifts last-hop recall on MuSiQue and 2Wiki without discarding magnitude information.

arXiv 2603.28886 →
Paper 2 · Multi-hop retrieval

BridgeRAG

For questions that need a chain of evidence, the second-hop document should be ranked by usefulness given the first hop, not by similarity to the original question. Training-free, graph-free, beats published baselines on three standard benchmarks.

arXiv 2604.03384 →
Paper 3 · Per-query routing

RegimeRouter

Different question types need different retrieval strategies. A lightweight router picks the right mode per query, with measurable gains in domain and graceful behavior out of domain. No hand-labeled query types required.

arXiv 2604.09019 →
Paper 4 · Controlling authority

CAR

Finding the currently valid document is a different problem from finding the most similar one, and the two metrics are formally decoupled. CAR validates the framework on FDA, SCOTUS, and security advisories with large gains over dense baselines.

arXiv 2604.14488 · GitHub
Early access

Pilots open with a small number of design partners.

Currently in active conversations with finance and legal teams across mid-market RIAs, regional banks, AmLaw firms, and large-cap controller offices. We onboard four to six design partners before commercial GA.

Design partner gets

Hosted endpoint, integration support, and a co-built ingestion adapter for one corpus.

Pricing

Design-partner license at a discount from list. Roadmap input. No long-term lock-in.

Timeline

Two-week scoping. Six-week integration. Twelve-week paid pilot decision.

Or write directly to andre@portmem.com.