Private beta

Know what your automation does β€” and where the evidence ends

Grounded, Honest and Continuous

Most automation drifts with time β€” people move on, logic gets changed, alarms and overrides pile up, and eventually no one has a verified picture of what it actually does. Daritas reads the real automation code and rebuilds that picture claim by claim: every claim cited to its source and graded by how it is known, the unknowns disclosed instead of papered over, reviewed and signed off by a named automation engineer, and kept current as the system changes.

Request beta access

Creates your account. We open it by hand β€” usually within a day, and nothing is charged.

Not ready for an account?

Just your email, no account and no login β€” we write when there is something worth reading.

Two ways in β€” start where you stand

Keeping the plant running well is a shared interest. The case for Daritas just looks different from the corner office and from the plant floor.

Whichever door you take, it isn't a one-off report β€” Daritas follows your system through change, version by version, and keeps the picture current as the plant evolves. See the living evidence loop ↓

Why it matters to the business

Engineers trust it because every claim is grounded in the real code. Leadership backs it because it de-risks change, protects uptime, keeps critical knowledge in-house, and rolls the whole estate up to a clear picture of what's sound and what needs attention.

See the whole estate at a glance

Every system across your sites, rolled up for leadership β€” what's sound, what needs attention, and where the risk sits. The executive view of your automation.

De-risk change & modernisation

Know exactly what to preserve, change, or retire before you touch production or brief an integrator β€” so upgrades don't turn into expensive surprises.

Fewer unknowns, less downtime

Undocumented logic and forgotten overrides are how production stops for reasons no one can explain. A verified baseline removes the blind spots.

Keep the knowledge when people leave

Retiring engineers and rotating contractors take the understanding with them. Daritas captures it as a durable, evidence-backed asset β€” evidence on demand for handovers and audits.

Your AI already answers questions about your automation code

Ask it anything and it answers. Knowing which questions matter is harder β€” and your own automation engineers have spent careers learning exactly that. What they haven't had is the countless engineering hours it takes to move that judgment from a brain to a keyboard: turning it into a repeatable, evidence-disciplined method an analysis can run on β€” that is what Daritas is. So when an experienced engineer is heading for the door, the best use of the months you have left with them isn't building an analysis workflow in an AI workspace β€” it's answering the sharp questions our assessment asks, so what they know becomes owned, confirmed knowledge of your installation before it drives away. And when a finding needs fixing, the assessment can brief the AI in your own engineering tool β€” the verified context it cannot see from the code alone: what the function is for, which interlocks the fix must not break, what to re-verify afterwards. Not a second AI to argue with yours β€” the evidence layer above it, giving the one you already use verified context and explicit uncertainty bounds, so acting on its answers carries less risk.

The answers are cheap now. The questions are the expertise.

When the code can't answer, Daritas asks the people who can

Some things were never written down: why a setpoint is what it is, which alarm everyone learned to ignore, the workaround from the 2019 retrofit. Where the sources can't settle a behaviour, Daritas doesn't fill the gap with a plausible sentence β€” it raises the question in plain language, quotes the line that prompted it, and names who is likely to know: the controls engineer, the operator on nights, the integrator who built it, the vendor's support desk. Bring your team into the portal and the open questions sit there against the model, one per behaviour waiting on a person. Every answer is recorded against the claim it explains, attributed to who said it and when, and kept as testimony β€” a separate tier from what the code proves, never laundered into fact. We ask the few questions that would change the model, never a questionnaire.

An unanswered question stays a stated unknown β€” never a guess.

Why engineers trust it

Under the hood: nothing is asserted without inspectable evidence β€” every claim grounded in its source, graded by how it is known β€” and in the expert lane, signed off by a named engineer.

From code to clarity

Thousands of lines of ladder logic and structured text, surveyed and distilled into a model you can actually read β€” in days, not the months a manual reverse-engineering takes.

Expert-reviewed

In the expert lane, a named, accountable automation engineer reviews the model and attests to its claims within a stated scope. Not a black-box AI summary β€” a result a human stands behind.

Vendor-neutral by design

Daritas reads the logic itself β€” across controller families, IEC 61131-3 languages, and embedded control code β€” not a single vendor's format. Support depth varies by format, and every assessment states what was fully parsed and what was not.

A living model

Every claim links to the evidence that supports it β€” source-derived claims down to their source locations β€” and the model follows the system version by version, instead of going stale in a drawer.

Ask it on the spot

Standing at the machine with an alarm you do not recognise? Ask, and get an answer drawn from that system’s assessed model with the evidence behind it β€” scoped to the machine in front of you, because the same code means different things on different systems.

Built for the age of AI in the control room

As AI starts to explain, change, and advise on production systems, it needs trustworthy context β€” not guesses. Daritas is the verified evidence layer: it shows what the evidence establishes about your system today, flags what's unknown instead of inventing it, and shows what changed after an AI-assisted change. When AI touches your automation, you keep the evidence trail.

Read-only. Advisory. Never a control path.

Ask your system a question. Get an answer it can prove.

Every assessment comes with the Advisor: ask in plain words and get an answer drawn only from that system's verified model, each source quoted with its review status and confidence. It is not a chatbot with your documents in the prompt β€” retrieval is deterministic, the model only phrases what was actually found, and when the evidence does not cover your question it says so instead of filling the gap.

βœ“Answers cite the behaviours they came from β€” file, location and the quoted line
βœ“Each source carries its own grade: machine-generated or engineer-reviewed, never blurred together
βœ“No verified evidence for the question? It says that, and names what it does have
βœ“Keep an answer private or share it with your organisation β€” your call, per answer
βœ“When the evidence cannot answer, we ask a named person β€” in chat, by email, by SMS, or a call to their desk β€” and their answer joins the model attributed to them

And when your material cannot answer, someone gets asked

The half of a Q&A engine nobody builds: the questions that go the other way. Where the evidence runs out, the engine raises a question instead of guessing and it is put to whoever holds that knowledge β€” through our own chat, email, SMS, or a phone call that tells them one is waiting. The answer becomes evidence with their name and the date on it. Nobody answers in time? The gap is recorded as a labelled assumption, and it stays labelled β€” that is the opposite of a confident sentence you cannot trace.

And it works as a team, not one person's chat window

Answers are kept against the system, so the question asked at three in the morning is still there next month β€” and the person who asked can share it with their organisation, turning one night's answer into something the whole team can find. Ask on the phone at the machine, read it back at the desk.

Read-only. Advisory. Never a control path.

Fast or expert β€” your choice, per occasion

Every deliverable carries its grade β€” machine-generated or engineer-reviewed β€” and the label is never blurred. The reviewed grade is worth paying for because it is attributable β€” bound to a named reviewer, a defined evidence set, and a stated scope.

The fast lane β€” machine-generated

An AI-only pass at machine speed and machine price β€” for breadth, orientation, and the everyday questions that don't warrant an engineer's time. Same grounding discipline, honest label: no engineer has reviewed it, and it never pretends otherwise.

The expert lane β€” engineer-reviewed

A qualified automation engineer verifies the model against the evidence, resolves ambiguities, and signs off by name. For the occasions where accountability matters: safety, compliance, handover, the modernisation decision.

Same customer, both lanes β€” breadth from the machine, accountability from a name.

Read the Assessment Standard β€” what we assert, and how to check it β†’

A living evidence loop β€” not a one-time report

Your installation changes: code gets modified, systems get modernised, people move on. Daritas keeps the understanding current. Every change gets a change-impact review, every unknown becomes tracked work, every baseline is versioned and signed off β€” a Plan–Do–Check–Act loop that turns a one-time assessment into a durable record, current to the latest accepted baseline: when a new version lands, affected claims are marked stale until their evidence is renewed.

That's the Ξ΄ in our mark β€” the engineer's symbol for change. An assessment done once tells you what your system was; used continuously, Daritas shows every delta: what changed between versions, with the evidence for it.

Portable markdown β€” yours to keep, even without the subscription.

From one controller to a system of systems

Real installations are rarely a single program. Daritas assesses multi-unit systems as one composed model β€” each unit read at the depth it needs, and every connection documented as an Interface Catalog (ICD): between units, and outward to SCADA, field I/O, and neighbouring systems. Proven behaviour and design-level structure are reported as separate tiers, so you always know which claims carry evidence.

Interfaces evidenced in the code β€” runtime behaviour marked unverified until corroborated.

Where it lands

Three stories from industry and one from the Moon β€” what an evidence-backed model does in practice.

How it works

Runs on your floor, with your own AI β€” a local AI model, or your API key over one controlled connection. The engine does the reading; in the expert lane a named engineer reviews the result and attests to its claims within a stated scope β€” an evidence-backed system model you can inspect, question, and keep current as the system changes.

01

On your floor, on your terms

Daritas runs inside your own network β€” point it at your control program (a PLCopen XML export or the project source). The deep reading uses an AI: your own local AI model, your API key over a single controlled connection, or β€” for fully air-gapped sites β€” the deterministic engine alone, with the reduced depth honestly marked. Your source never has to leave your site.

02

Daritas surveys & extracts

The engine surveys the code, condenses it to its logic essence, and extracts the behaviours it can identify β€” each tied to the exact source that supports it, with coverage disclosed: what was read, what was out of reach, what stayed unknown. Bring your own AI: a local AI model or your own API key, so even the reading stays under your control.

03

An engineer verifies & signs off

A qualified automation engineer checks the model against the evidence, resolves ambiguities, and puts their name to it.

04

Explore your model

Get an interactive model you can navigate, plus clean document exports for your own meetings, audits, and handovers.

Security & data

Built for control rooms that can't send their source anywhere. You stay in control of what gets read, what gets shared, and what you keep.

Air-gapped

The engine runs offline, inside your own network. Your control-system source never has to leave your site.

Bring your own AI

Use a local AI model or your own API key β€” or, for the strictest sites, run the deterministic engine with no external AI at all.

No black box

Every claim links to the evidence that supports it, and reviewed models carry a qualified engineer's sign-off. Nothing is asserted without evidence.

You own the record

Only the evidence-backed model leaves, and only if you choose. It exports as portable markdown β€” yours to keep, even without the subscription.

Your IP stays yours

Review and redact before anything is shared, and optional obfuscation keeps proprietary logic out of what you hand on. Daritas never uses your source to train models β€” and with a bring-your-own provider, the training terms are the ones in your own contract.

Isolated by default

In the cloud, your system model is locked to your organisation β€” row-level security, access by invitation, and access tokens stored hashed.

For system integrators & consultancies

If you deliver and maintain automation for others, Daritas is your leverage β€” assess and verify your clients' systems with your own team, under your own brand, across every site.

Your brand

Hand over the evidence-backed model and reports as your own team's work β€” the Daritas branding is removable.

Your team, no external eyes

Verify with your own engineers β€” juniors do the volume, a senior signs off. Nothing leaves your client's site.

Every client from one place

Switch between client sites and see the whole estate β€” one login across all the organisations you serve.

Priced by depth of assessment

Start with one system. Buy the baseline once, keep it living with a continuity plan, and expand across your estate β€” priced by scope and criticality, never per seat.

Each assessment is scoped and quoted before you pay β€” you see the price after the scoping preview, and nothing starts until you confirm. Pricing confirmed on enquiry during early access.

High-level map
Orientation
Priced at scoping

A fast, high-level map of one system: its main components, structure, and most visible risks. The cheapest way to decide where to dig.

Identified behaviours, grounded
Full model
Priced at scoping

Every identified behaviour grounded to its source, with components, dependencies, and a risk picture β€” plus a coverage report of what was analysed, what was not, and the open unknowns. The core deliverable for a system you depend on.

Most popular
+ sceptical review pass
Verified
Priced at scoping

A full model plus a deeper, adversarial review β€” does the evidence really support each claim, do claims contradict each other, and what is obsolete. For audits and high-stakes systems.

By agreement
Advanced
Priced at scoping

Verification against a replacement, a specification to rebuild from, or runtime corroboration. For replacement, modernisation, and verification programmes.

Start small, own the baseline

Begin with a scoped Evidence Sprint β€” a narrow, fixed-price slice that proves the value and credits toward the full assessment. Then own the baseline at a fixed, quoted price per system. No subscription required for your first model.

Keep it living

Subscribe to keep the model current as the system changes β€” re-runs and diffs that keep it the system of record instead of letting it go stale.

Expand across systems

Add more systems, sites, and variants over time β€” and get one comparable overview across your whole portfolio.

</>Available now β€” read API & MCP

Your system model. Programmable.

Every assessment builds a structured, queryable model you own. Reach it via API, an MCP server, or a documented export β€” so your system knowledge lives inside your own IT, your own tools, and your own documents.

Queryable model

Every delivered system model lives on as a queryable dataset. Ask in plain language β€” "which interlocks stop the supply fan?" β€” and get structured answers with source citations, not a PDF page number.

MCP server

Connect Claude, Cursor, or your own AI agent straight to your system model. Let it look up behaviours, trace dependencies, and check claims against the evidence. With a local AI model the loop stays inside your network; with a cloud AI client, tool results reach that provider under your contract.

Yours to take with you

Every assessment exports in portable form β€” Markdown, Word, PDF, and a structured export of the model itself. No lock-in: the understanding of your own plant leaves with you, readable without us.

# Read your model over the API, with your own key
$ curl -H "Authorization: Bearer $DARITAS_KEY" \
https://daritas.com/api/v1/assessments/ahu-03/behaviours
The supply fan stops on fire alarm, low-limit frost, or smoke detection.
FB_SupplyFan.st:142 Β· FB_Safety.st:88 Β· IO_Map.st:31
# Or let Claude / Cursor call it as a tool over MCP
get_behaviours(assessment_id: "ahu-03")

A named engineer behind every reviewed model

Daritas isn't a black box. Every reviewed model is verified and signed off by a qualified automation engineer who puts their name to it β€” because in industrial control, accountable human judgement beats a confident guess. We're building a global network of those engineers, with layered AI and peer checks that reduce the risk of a single weak review. For sensitive systems, an optional obfuscation layer strips your proprietary names before anyone sees the code, then restores them in the delivered model β€” your logic is reviewed, your proprietary names stay home.

Join the reviewer network