Kognitos and Lleverage both automate work without writing code, by opposite routes. Kognitos asks you to describe the process in plain English, then executes that description deterministically. Lleverage builds agents around a specific back-office process and runs them inside your ERP. In our opinion the choice comes down to who owns the work after go-live.
Both are worth taking seriously, and they are not competing for the same buyer as often as the category labels suggest. Kognitos sells into large enterprises with a process-automation function and a mandate to consolidate. We are usually called into a manufacturer or wholesaler of a few hundred people, where the order desk is the constraint on growth. That is the vantage point this comparison is written from.
If you are weighing the two for order intake or invoice matching, the fastest way to settle it is your own documents. Book a demo and bring the messy ones.
What is the difference between Kognitos and Lleverage?
Kognitos is a general-purpose automation layer where the process is written as English sentences and executed deterministically. Lleverage is narrower: agents built for the back-office processes of companies that make, move and sell physical products, running inside the ERP those companies already own. One is horizontal by design. The other is deliberately not.
| Kognitos | Lleverage | |
|---|---|---|
| How a process is defined | Written as plain English statements | Built around a named process with your rules and exceptions |
| Execution model | Neurosymbolic: symbolic logic with language understanding on top | Agents acting over ERP and system interfaces |
| Primary buyer | Large enterprise, often a central automation function | Operations and finance leads in SMEs that make, move and sell products |
| Industry focus | Horizontal: finance, healthcare, banking, supply chain, IT | Manufacturing, wholesale and distribution, logistics, process industry |
| Published proof | Enterprise document and finance operations | Order intake, invoice matching and support inside named European operators |
| Data residency | US-headquartered, enterprise compliance certifications | EU hosted, GDPR |
| Published pricing | No public tiers, entry tier available on request | Per agent per month, from 2,000 euros |
| Implementation | Build it yourself or with a partner | Forward-deployed: our engineer ships the first process with your team |
The row that matters most is the first one. Kognitos treats the English description as the program. That is a genuinely different idea from prompting a model, and it is the source of both its strengths and its limits.
What does Kognitos actually do well?
Kognitos executes plain-English process descriptions deterministically. The same input produces the same result, and every step is logged. For an enterprise that has been burned by probabilistic automation that is a serious answer to a serious objection, and the published customer results are not pilot numbers.
The architecture is the interesting part. Rather than asking a language model to decide what to do at runtime, Kognitos compiles the English into symbolic logic and runs that. The model handles understanding, not execution. The company calls this a neurosymbolic approach, and launched on that positioning alongside a 25 million dollar Series B announced in June 2025, led by Prosperity7 Ventures. Total disclosed funding stands at roughly 78 million dollars.
The proof they publish is enterprise-scale. More than 50,000 bills of lading are processed monthly at Century Supply Chain Solutions. Around 3,300 hours of manual work a year were removed at JBI Interiors. A Fortune 50 manufacturer is cited with a projected 23x return. HFS Research named the company a Hot Tech 2026 pick on the strength of the deterministic execution and the audit trail.
There is a second thing they do well that gets less attention. Because the process is written in English, the person who understands the process can read it. In most automation projects the business description and the running implementation drift apart within a quarter. Keeping them as one artefact answers a real problem, and we would not pretend otherwise.
Where does natural-language automation stop short?
Plain English describes a process cleanly, right up to the point where it meets a twenty-year-old ERP and a customer who sends orders as a photograph. What is hard in back-office work is rarely the logic. It is the mess at the edges, and that mess is specific to your systems and your customers.
Three limits show up repeatedly when we are called in after a horizontal product has been tried.
The first is that a general-purpose layer has no opinion about your domain. It does not know what a sales order means in a wholesale business, what tolerance a supplier invoice should clear at, or what to do when an article code is unrecognised. You supply all of that in English. You keep supplying it as reality changes. That is work, and it lands on whoever wrote the description.
The second is the last mile into the system of record. Reading a document is half the job. The other half is posting a validated order into Business Central: the right customer, the right article codes, the right pricing rules, the right confirmation back to the sender. That is the half that decides whether the desk's capacity actually changes. Horizontal products connect at the surface and hand the finishing back to a person.
The third is the buyer profile, which is not a criticism so much as a fact about fit. Kognitos is built for organisations the size of Paysafe and PepsiCo, with an automation function to run it. A 200-person wholesaler has an operations lead and an ERP consultant on retainer. It has no capacity to own a general-purpose automation layer. Our view is that this, rather than any feature gap, decides most of these evaluations.
"The answers come back with references to the exact manual and section. That is what builds trust — the technician can verify the information themselves before passing it to a customer." – Remco Hooft, Technical Owner, Oude Reimer
How does Lleverage handle the same work?
Lleverage builds an agent around one named process and runs it inside the customer's existing ERP. What it cannot settle goes to a person with the context attached. The agent is not described in English. It is built with your team, corrected on live documents, and widened only when the log shows it holds up.
In practice the difference is visible in the first fortnight. Rather than writing a description and testing it, the agent reads your actual inbound documents while a person approves every result. The corrections from that period are the product. They encode the customer who uses a different unit of measure, the supplier whose reference number needs reformatting, the article code that maps to two SKUs. Xpol's intake agent codifies 25 customer-specific rulesets that previously lived in the heads of senior staff.
The results we publish are from that shape of work. Topa Bathroom Products posts more than 90% of inbound orders straight into Business Central, with confirmations back to the customer inside 30 seconds. Close to four full-time roles moved onto after-sales. At Oude Reimer, technical triage that once meant scrolling through hundreds of pages of German manuals now returns a referenced answer in 70 seconds.
None of this makes ours the broader product. It makes it the narrower one, on purpose. Our agents run inside the systems our customers already have, which is a smaller claim than automating anything in any industry, and a more useful one if you are the person who has to sign off the month.
Which one fits a mid-sized manufacturer or wholesaler?
For a company of a few hundred people running Business Central, SAP, Exact or AFAS, where the constraint is an order desk or an AP queue rather than a central automation programme, we think Lleverage is the better fit. For a large enterprise consolidating many automation products under one governed layer, Kognitos is a credible answer. We would say so in the room.
The honest test is not a feature comparison. It is three questions about your own organisation.
- Who owns this on the Tuesday after go-live? If the answer is a central automation team with capacity, a horizontal layer works. If it is the operations lead who also runs the desk, a product that arrives already shaped to the process will survive. A general-purpose one usually will not.
- Does the work have to finish inside the ERP? If success means a record posted, matched and confirmed, weight ERP-native execution heavily. If success means a decision surfaced to a person, the requirement is lighter and more products qualify.
- How specific are your exceptions? Generic processes suit generic products. If your desk runs on twenty-five customer-specific rules that nobody has written down, the value is in capturing those rules. That is a different exercise from describing a happy path in English.
That is where we land, and it is a judgement rather than a measurement. Run both against the same week of real documents, including the ones that usually go wrong. The answer tends to settle itself.
What does each cost?
Kognitos does not publish pricing tiers. An entry tier is available on its site, and everything beyond that is a sales conversation. Lleverage publishes its pricing: one monthly price per agent, from 2,000 euros a month for a standard mid-market deployment. Integration, AI usage and ongoing improvement are included rather than metered.
The comparison worth making is not the headline number. It is what the number covers and what it excludes. Per-seat and per-consumption models make the second and third process cost roughly what the first did. They also make the bill move with volume, which is exactly the volume you were trying to grow. Pricing per agent ties the cost to a job rather than to usage. Our own view is that an agent doing a job should sit on the same budget line as the people doing that job today.
The other cost that rarely appears in a comparison is who writes and maintains the description of the process. With a horizontal layer that is a permanent internal role. With a forward-deployed model, the vendor's engineer builds the first process with your team and hands over something already running. Both are legitimate. Only one of them shows up on the invoice.
Frequently asked questions
Is Kognitos a competitor to Lleverage?
Only partly. Both automate back-office work without traditional development, so they appear on the same shortlists. In practice Kognitos sells a horizontal automation layer to large enterprises. Lleverage sells agents shaped to specific processes in SMEs that make, move and sell physical products. The overlap is real but narrower than the category suggests.
Does Kognitos work with SAP or Business Central?
Kognitos integrates with major enterprise systems including SAP, Oracle, Microsoft Dynamics and Workday, according to its own site. The question to ask any vendor is not whether a connector exists but how deep it goes. Reading from a system is common. Posting validated transactions back into it under your approval rules is the harder half.
What does deterministic execution actually mean?
It means the same input produces the same output every time, because the logic runs as code rather than being decided by a model at runtime. Kognitos uses language understanding to interpret inputs and symbolic logic to execute. It is a sound answer to the reliability objection, and it does not by itself solve domain fit or the last mile into an ERP.
Which is faster to get into production?
It depends on who is doing the building. A horizontal layer can produce a first working process quickly, if you have people who can describe processes precisely and maintain them. A forward-deployed model puts a vendor engineer on the first process. That is usually faster for a company without an automation function. Both take weeks rather than months for one well-chosen process.
Where is the data hosted?
Lleverage is EU hosted and GDPR compliant, which matters to European buyers with customer and supplier data in scope. Kognitos is US-headquartered and publishes SOC 2, ISO 27001, HIPAA and GDPR compliance badges. If EU data residency is a procurement requirement rather than a preference, confirm it in writing before a trial.
Try both against the same week of documents
Comparisons like this one settle in an afternoon if you use real inputs. Take one week of inbound orders or supplier invoices, including the ones your team normally fixes by hand. Put them through each product end to end, and count what finishes in the ERP without a human touch. That is the only number that matters.
We are happy to be measured that way. Book a demo with your own documents. Or read how we draw the line between supervised agents and background automation before you scope the first process.