Verifiable ERP AI for Infor M3 and CloudSuite

The only ERP AI that will refuse to answer.

Dovrin is the grounding gate that sits between AI and your ERP data: it cites every number back to the source, or it withholds the brief. Infor is giving your teams agents. Dovrin is the control that makes them safe to put in front of a CFO or an auditor.

Every number shows its receipts. The ones it can't, it won't say.

Built for Infor M3 CloudSuite Tenant- & ERP-agnostic core Truth Protocol, every claim source-tagged Own failures published, dated, fixed
The gap

Enterprise AI can advise. It can't be trusted to act.

The market is full of copilots that read your ERP and suggest. The last mile, actually changing a credit limit, a BOM, a reorder point, is where trust breaks down. Dovrin is the governed layer that closes it.

AI today

Observes & advises

  • ✕  Answers from training data, no proof it's true for your tenant
  • ✕  Suggests a change, but a human still does the work by hand
  • ✕  No view of what a change breaks downstream
  • ✕  If it does write, it's a black box, no audit, no undo trail
AI with Dovrin

Safely acts, under governance

  • ✓  Every fact carries its source; unverified claims are withheld, not guessed
  • ✓  Proposes a real ERP write as a draft, never executes on its own
  • ✓  Shows the consequence of the change before anyone confirms
  • ✓  A human confirms; the action lands once, in a tamper-evident audit
See it refuse

It won't lie to you. Watch it prove that first.

Before Dovrin ever writes to your ERP, it reads and narrates. The same Truth Protocol gates both halves: state only what the data can prove, and say so out loud the moment it can't.

Dovrin Narration  ·  Acme Distribution  ·  Accounts Payable
Request
◆ AP Manager
"What's overdue right now, and should I be worried?"
Withheld, not guessed
1 invoice came back for this pull. Its latest status record doesn't clear the confirmed-open bar, so Dovrin won't state a balance it can't cite.
·canonical status unconfirmed on the latest record
·no corroborating read to confirm the balance
·confidence below the threshold to narrate
This is Dovrin working as designed, not a failure state. A confident wrong answer costs more than an honest "not yet."
The most valuable thing an AI can tell you is: "I can't verify that yet."
Interactive demo · representative data
Grounded Narration
Ask a question above. Dovrin narrates only what it can cite, and says so the moment it can't.
Dovrin narrates only what it can cite. Thin or ambiguous data gets an honest refusal, not a guess.
See it work

A governed write, start to finish.

An agent proposes raising a customer's credit limit. Watch Dovrin ground it, preview the consequences, and hold it for a human, the way a real change should happen.

Dovrin Console  ·  Acme Distribution  ·  CONO 100 / DIVI 001
Request
▴ Buyer agent · @o2c
"Customer C100 keeps hitting their credit ceiling. Raise their credit limit to $150,000 so their orders can ship."
Proposed write Draft · not executed
program  CRS315MI txn      ChgCreditLimit CUNO     C100 CRLM     50000 → 150000
Interactive demo · representative data
Consequence Preview
Before anyone confirms, Dovrin shows what this change will do downstream.
Dovrin executes only on human confirmation. Every fact above is source-tagged; nothing is asserted that wasn't read.
The governed loop

Five steps. No write skips one.

Dovrin turns "the AI changed something" into a sequence you can defend in an audit. The same gate holds on every path, chat, agent, or integration.

1

Ground

Read live ERP state. Every value is tagged with its source and confidence.

Truth Protocol
2

Draft

The agent proposes a real ERP write, as a draft. It holds no power to execute.

advisory-only
3

Preview

Show the downstream consequences before anyone commits, orders, exposure, policy.

Consequence Preview
4

Confirm

A verified human approves. The agent never holds the confirmation key.

human-in-the-loop
5

Audit

Execute exactly once. Record it in a hash-chained log that detects any tampering.

tamper-evident
Why it's trustworthy

Governance isn't a feature. It's the foundation.

Most AI tools bolt on guardrails. Dovrin is built inside-out from three rules, and unlike a deck, two of them come with a dated receipt.

Pillar 01

Every claim is sourced, or it's withheld. Never guessed.

Source tiers are enforced in code, not asserted in a deck, seven tiers from live-tenant confidence down to UNVERIFIED, which is a hard stop. The answer is withheld, not degraded. The same discipline runs in the free H5 Scripting Kit, which tags every claim [VERIFIED] or [CONFIRM ON TENANT].

Pillar 02

Every write is human-reviewed before it touches your tenant, with a record that can't be quietly edited.

A draft is created carrying no token; a human approves at a dedicated confirm surface, which mints a one-time token server-side, an agent can never self-confirm because it never holds one. Consequence Preview shows the downstream impact first, so the human approves an outcome, not a field diff. The result lands in a tamper-evident, hash-chained ledger.

Pillar 03

When our own governance is broken, we say so, in writing, dated, before a customer finds it.

The pillar competitors can't copy with better marketing, because it requires publishing your own failures.

2026-06-10 A commissioned adversarial audit found the write gate's advertised "triple lock" was, as wired by default, a double lock. Found offline, before any tenant. The standard's own changelog now states 2 of 3 locks are enforced and the third is deliberately NOT_ENFORCED and loud, a required literal with no silent permissive default.
2026-08-30 A triple-pass verification of the whole portfolio found the first pass had undercounted its own test suite by roughly 13%, and caught a deterministic CI failure on main no prior build had surfaced. Fixed the same day; suite green.
How the pillars are enforced

The mechanics behind the three rules.

Truth Protocol

Every claim about your ERP carries its source and a confidence score. Unverified claims are flagged and never clear the bar to write, the system is built to surface "I can't verify that" rather than pass a guess off as fact.

Human-confirmed writes

An agent can propose a write, never execute one. A verified operator confirms; the action runs exactly once. No agent ever holds the key.

Consequence Preview

Before a change is confirmed, see what it touches downstream, held orders, open POs, credit exposure, each impact traced to the live read it came from.

Tamper-evident audit

Every operation is hash-chained, recording what the human saw at confirm. Any edit, deletion, or reorder is detectable, built for the auditor.

Agnostic by design

One governed core, any tenant, any Infor product. M3 is the first adapter, not the foundation. The same trust layer extends to LN and beyond.

Powered-by model

Dovrin is the engine inside your ERP automation, "<ERP> Agentic Solutions, powered by Dovrin." Your workflows, our governance underneath.

The lineup

Two products. One discipline.

Not seven SKUs, a flagship you open every day, the migration accelerator that proved the discipline first, and a set of free tools that show the same craft before anyone talks to sales.

Flagship · the hero

Dovrin Toolbox

A daily-use dashboard for any M3 ERP product, ticketing integrations, working ION API knowledge, and AI, with Consequence Preview and grounded narration built in. This is "M3 Agentic Solutions, powered by Dovrin."

Narrate, cited briefs Consequence Preview Human-confirmed writes
See the Toolbox → See narration live →
Second product · the receipts, shown early

Dovrin Data Accelerator

Synthetic-tenant generation and governed load for CSD/CSDE migrations, the most mature build in the portfolio, replay-gated end to end. Migration is a project, not a daily habit, but it's the fastest way to see the rigor before trusting the Toolbox.

Synthetic-tenant fleet Migration corridor Replay-gated CI
See the Accelerator →
The governed rail underneath both products, containerized create → rehearse → verify, with a tamper-evident hash-chained audit, is built and gated (173 tests, image-gated) but wired to a mock adapter only. It's cited here as why the audit trail is credible, not pitched as a separate product; live-tenant wiring is the next milestone.
Free, on the site

Not ready to talk to sales? Start here.

Small, single-purpose tools, low commitment, same craft-signal as the flagship.

Lead magnet

SOW Cost Estimator

A live SOW cost, price, and margin estimator with hard margin-floor guardrails and versioned rate cards.

Try it →
Lead magnet

M3 Activation Tracker

A repeatable, engagement-agnostic go-live playbook for CSDE, two parallel tracks, gated phases, live progress.

Try it →
Lead magnet

H5 Scripting Kit

A practitioner kit for M3 H5 panel scripting, starter recipes and docs, every claim tagged verified or confirm-on-tenant.

See the kit →
One core, every ERP

M3 is where we start, not where we stop.

The governance is product-agnostic by design. The same gate, Consequence Preview, and tamper-evident audit ride on top of any Infor ERP's connector, M3 is the first adapter, not the foundation.

Built · offline

Infor M3

The first adapter. The full governed loop runs against M3 today on representative data, order-to-cash, procure-to-pay, and manufacturing.

Adapter scoped

Infor LN

Rides the same ION OAuth backbone, so the governance core reuses directly. Connector researched and designed; live work awaits a partner tenant.

On the roadmap

CloudSuite Industrial (SyteLine)

A different API model under the hood, the same trust layer on top. Sequenced after LN.

By design

Your ERP

The gate, preview, and audit assume nothing about M3. Bring a connector; the governance comes with it, "<ERP> Agentic Solutions, powered by Dovrin."

Where it lands first

Built for the most-audited corners of the ERP.

The higher the stakes of a bad change, the more a governed write is worth. We start where a mistake is a recall, a stockout, or a blown margin.

Flagship

Food & Beverage

Recipe, allergen, shelf-life and lot changes, preview the blast radius before a single record moves.

Live now (mock)

Manufacturing

BOM, routing, reorder-point and production-order changes, see the open orders a change re-drives.

Foundation

Distribution

Credit limits, pricing, item blocks, supplier banking, the order-to-cash and procure-to-pay core.

Design-partner program

Put governed AI in your M3, with a partner, not a pilot light.

We're onboarding a small number of design partners to run Dovrin against a real tenant. You bring the M3 environment and the use case; we bring the governed engine and build it with you.

Honest status: Dovrin is design-partner-ready, not yet generally available. The governed loop runs end-to-end on representative data; live-tenant validation is exactly what the design-partner program is for.