Public beta, live and self-serve · Design partner program open, finance first

Search, build, and check AI work against the documents that govern it.

And keep the proof.

PortMem checks every claim before it ships: what's wrong gets flagged, what's missing gets named, and every step lands in a signed, tamper-evident ledger with human sign-off.

Test it today See the live demo

Cost

You pay for the passages, not the pile. Agents read what matters, not 500,000-token pastes, so cost tracks what you retrieve and falls as models get cheaper.

Speed

Minutes, not review cycles. Reviews run while the analyst is still reading, not overnight.

Traceability

Every step on the record. Verdicts and sign-offs land in a signed, tamper-evident ledger.

How it works

Submit a report. Get the audit.

Upload the governing documents; versions and amendments handled.

Submit the AI-written report, from any tool.

Get the audit — per-claim verdicts, evidence quotes, omissions, human sign-off.

Export the proof — a signed evidence packet an examiner can verify offline.

PortMem review console: delivery gate clear, contradicted claims with evidence and human sign-offs

The production console: real verdicts on public deal documents, verified by hand against the deed text. An agent can run the whole loop over MCP. See the live demo →

Q: "Is ASC 605 still the controlling revenue standard?"

Standard RAG ASC 605 governs revenue recognition under US GAAP, covering… × cites stale guidance; misses the 2018 supersession.

PortMem No. ASC 605 was superseded by ASC 606 (ASU 2014-09), effective fiscal years beginning after Dec 15, 2017. ✓ signs the verdict.

The solution

One review, four guarantees.

Claim-level verdicts.

Every claim in a report cites the exact passages it derives from, in the governing version of the document. A citation that is faithful to a superseded contract gets flagged, not certified.

Omissions named.

Contradicted claims are caught before the report ships, and omissions are named: the provisions the document requires that the report never covered. Both directions, wrong and missing.

Signed evidence.

Every verdict and sign-off lives in a hash-chained, Ed25519-signed, append-only ledger. Anyone holding our published key can verify our records. Nobody holding it can forge one.

The version that governs.

Answers come from the version that currently governs. Superseded, amended, and recalled documents are filtered out at retrieval, not cited as if they were still in force.

Built on four published arXiv papers: PhaseGraph · BridgeRAG · RegimeRouter · CAR · How we compare to every adjacent stack →

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

95% Same model, no documents

Across FASB Accounting Standards Updates, the latest superseding document wins. PortMem picks it deterministically, with the citation and the controlling version — the baseline answers from memory with neither.

Legal

Has this precedent been overruled?

99% PortMem

48% Same model, no documents

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

29% Same model, no documents

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

Human-adjudicated runs on deal documents and three unseen regulated domains produced zero false contradictions across ~210 reviewed claims and caught every planted value error. The one miss was published, fixed, and re-verified.

Ledger verification key: 506afc81aa39331a4c704a97bca26100a0d857a682a6fa2dd7274b0dae547c3e · The legal corpus was audited against primary sources and relabeled, with the audit published · Baseline: the same Claude Haiku 4.5 answering with no documents · Full protocol, artifacts, and our published corrections: Methodology →

Where PortMem fits

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

Generators grade their own homework; faithfulness scorers emit an opinion, not evidence; archives store without verifying. PortMem is the independent verification layer: claim-level verdicts, completeness, a signed ledger, and version governance in one system.

← swipe to scroll →

Approach	Currency	Audit	Multi-hop	Model-agnostic	Regulated	Speed	Cost
Long-context LLMs Sonnet 1M class	✗	✗	~	✗	✗	Slow	$$$
Naive RAG LangChain, LlamaIndex	✗	~	✗	✓	✗	Fast	$
Agentic RAG LangGraph, CrewAI, AutoGen	✗	~	✓	✓	✗	Slow	$$
Compilation-stage knowledge Pinecone Nexus, PageIndex	✗	~	✓	✓	✗	Med	$$
Vertical legal AI Harvey, CoCounsel, Hebbia	~	✓	~	✗	legal only	Med	$$$
Enterprise search Glean, Sana	✗	~	✗	✓	✗	Fast	$$
Vector store + rerank Vectara, Pinecone, Cohere	~	~	~	✓	✗	Fast	$$
PortMem The verification layer	✓	✓	✓	✓	✓	Fast	$

✓ Built and load-bearing ~ Partial or workaround ✗ Not built

The engine underneath

Retrieval built for regulated documents.

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.

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.

Calibrated abstention.

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

Multi-hop by construction.

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

Per-query routing.

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

Drops onto what you already have.

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

Research foundation

Four arXiv papers. One company.

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

Paper 1 · Calibrated Fusion

PhaseGraph

PhaseGraph fuses vector and graph scores on a common rank-based scale, lifting last-hop recall on MuSiQue and 2Wiki.

arXiv 2603.28886 →

Paper 2 · Multi-hop retrieval

BridgeRAG

Ranking the second hop by usefulness given the first, not question similarity, beats published baselines on three benchmarks.

arXiv 2604.03384 →

Paper 3 · Per-query routing

RegimeRouter

A lightweight router picks the right retrieval mode per query. 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.

arXiv 2604.14488 → · GitHub →

Early access

Pilots open with a small number of design partners.

The design partner program opens with finance teams: mid-market RIAs, regional banks, and large-cap controller offices. Legal, pharma, and security follow. We onboard design partners in a small first cohort 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.