Iris.ai

Positioning

Iris.ai is a ten-year-old Norwegian AI company that has quietly repositioned itself away from the individual researcher and toward the regulated enterprise. It was founded in 2015 near Oslo (in Stabekk) by Anita Schjøll Abildgaard, Jacobo Elosua, Maria Ritola and Victor Botev, raised a €7.6 million Series A along the way, and spent its early years selling tools to academics — the Researcher Workspace (RSpace) launched in March 2022 as a suite for mapping topics, screening documents and extracting structured data from scientific text. Visit iris.ai today and you will find none of that consumer framing. The homepage now reads "AI Knowledge Foundation for Regulated Enterprises" and sells to manufacturing, life sciences, energy and telecom teams running AI over their own documents and systems.

The core claim has narrowed into something more defensible and harder to demo: the system controls which sources a model may read and measures every step of how it reaches an answer, so outputs in high-stakes environments can be traced and defended. That is a provenance-and-control pitch, not a chat-with-papers pitch, and it explains why the buying motion is now request-a-demo rather than self-serve sign-up. Named customers visible on the site at time of writing include Syngenta, the USDA and Standard Industries.

This is the review in this batch I could not test hands-on, and I will be explicit about that throughout. Iris.ai has no public free tier, no self-serve trial and no published prices — navigating to iris.ai/pricing returned marketing content rather than a price table when I fetched it, and the buy-a-seat flow simply does not exist. Everything I observed came from the public materials, the product documentation I could reach, the company's published history, and direct comparison against tools in this category I have tested extensively. I think that constraint is itself the most important consumer fact about the product, so I have treated it as a tested observation rather than burying it.

What it does

The current offering, based on the vendor's own description, sits between an enterprise's document estate and whatever AI applications or agents the company runs. In practical terms the pieces I can identify: ingestion and structuring of scientific and technical text (papers, patents, internal documents) at scale; a controlled retrieval layer where experts define what the model may read; an answer path that records how each output was derived so it can be audited; and tooling carried over from the RSpace lineage — topic mapping (the old Science Map family of features), systematic screening of document sets, and structured extraction into comparable formats. The Researcher Workspace version history I read through shows this lineage clearly: iterative releases through 2022 and 2023 adding extraction, filtering and workspace features, then the pivot to the knowledge-foundation framing.

The vendor's positioning statement — that the system "controls what a model reads and measures every step of how it answers, so the result holds every time" — is a vendor claim and a strong one; "the result holds every time" is not something any AI system I have tested can promise, and I would push on exactly this sentence in a sales evaluation.

My testing notes

Observation one: the pricing opacity is total, and it is deliberate. I checked the pricing path, the sign-up flows and the third-party directory record for this review. There is no published price, no trial, and no free tier; a third-party directory describes Explorer, Researcher and Enterprise as quoted tiers with onboarding that requires data-engineering scoping across systems like ERP, SharePoint and document management. Third-party price aggregators float figures around $70 to $279 per user per month, but these conflict with each other and with the batch data I was given ($200 minimum), and none are corroborated by the vendor. I therefore state no price as fact in this review. For a buyer, the practical translation is: budget for a procurement cycle, not a credit-card sign-up.

Observation two: the public product surface has contracted. When I worked with Iris.ai's Researcher Workspace materials, the product was demonstrable — topic maps, screening workflows and extraction were visible in the vendor's own release notes and use-case write-ups (food-safety reviews, technology landscaping). Today's site leads with the knowledge-foundation layer and pushes everything else behind a demo request. That is a coherent move upmarket — regulated industries pay for control and provenance — but it means an independent evaluator cannot verify the interesting claims before engaging sales. I compared the workflow descriptions against what I know of comparable systems (controlled retrieval plus citation tracing is roughly what careful teams build by hand with open-source RAG pipelines plus an audit log); Iris.ai's differentiation, if it holds, is that this comes assembled with scientific-text tooling rather than as a construction project.

Observation three: the company signals are genuinely good. A decade of operation, institutional funding including European Innovation Council backing, named enterprise customers in exactly the verticals it targets (Syngenta in agrochemicals, USDA in government research, Standard Industries in manufacturing), and a technical blog that dates back years with real engineering content rather than marketing filler. In an AI vendor landscape where most companies are two years old and one model update away from irrelevance, Iris.ai reads as durable. The G2 and directory footprint is thinner than its age would suggest, which is consistent with an enterprise-sales customer base that does not write public reviews.

What it does not do

No self-serve anything: no free tier, no trial, no published pricing, no public sandbox. You cannot evaluate the product's actual output quality without a sales process, which is the single biggest limitation for a reviewer and for any small buyer. It is not a student or individual-researcher tool in its current form — the onboarding premise is that someone wires it to your enterprise document estate.

On the technical side, I cannot verify the accuracy claims (the "result holds every time" framing suggests near-deterministic behaviour that I would want benchmarked on my own document set before believing). Setup is described by third parties as requiring data-engineering work, which matches the enterprise positioning but contradicts any notion of quick deployment. And the product's scope is text understanding — it will not run your lab, do statistics on extracted effect sizes, or replace a full systematic-review process with human screening and inter-rater checks.

Verdict

Iris.ai is the most credible enterprise answer I have seen to the question "how do we run AI over scientific and technical text without losing the thread of where every answer came from?" — a decade-old company with marquee customers, real text-understanding lineage from the RSpace years, and a control-and-provenance framing that matches what regulated buyers actually need. But I could not test it, you cannot test it without a sales conversation, and the vendor's strongest claims ("the result holds every time") are exactly the kind that demand benchmarking on your own corpus before signature. I rate it 3.6 on the ToolSpotter rubric: core functionality depth looks substantial but is unverifiable from outside, value cannot be assessed without a quote, usability is enterprise-setup-class rather than self-serve, the support and company maturity signals are strong, and the longevity and funding record is reassuring. For an R&D or innovation team in life sciences, manufacturing or energy with a document-engineering budget, it is worth the demo; for anyone below enterprise scale, this is not your tool.

Sources

• https://iris.ai/
• https://iris.ai/pricing/
• https://iris.ai/blog/iris-ai-launches-the-researcher-workspace-ai-powered-tool-suite-improving-research-document-processing-for-r-and-d
• https://iris.ai/blog/announcement-the-researcher-workspace-1-0-is-finally-here
• https://iris.ai/blog/use-cases-for-the-researcher-workspace
• https://siliconcanals.com/iris-ai-raises-7-6m
• https://grokipedia.com/page/Irisai
• https://neuralcatalog.com/company/iris-ai
• https://aibestnav.com/ai-tools/iris-ai
• https://recatools.com/ai-directory/iris-ai