Iris.ai

A one-sentence summary

Iris.ai is an AI platform that converts papers, patents, and corporate documents into knowledge that can be searched, extracted, and verified; it currently serves research and development teams, knowledge management departments, and industries subject to regulation.

Tool Introduction

Iris.ai is operated by the Norwegian company Iris AI AS; it initially attracted the attention of researchers as a tool for discovering AI-related literature. Today, its role has evolved to that of an enterprise-level AI knowledge platform, utilizing Axion, Neuralith, and RSpace to handle data structuring, as well as Agentic RAG and research workflows.

Therefore, it can no longer be simply described as a website for searching papers. The module related to literature, which is of interest to individual researchers, remains part of RSpace; whereas corporate clients can integrate their internal documents, charts, patents, and databases into a more comprehensive system of knowledge and agents.

Three product positioning strategies

Structuring Axion data

Axion is used to convert knowledge scattered across documents, images, scans, tables, and graphics into a unified, machine-readable format. It does not merely perform OCR; rather, it creates a reusable data layer by taking into account context, metadata, entity relationships, and industry-specific schemas.

• Connects PDF, Word, Excel, PowerPoint, plain text, web pages, and scanned images.
• Identify values, labels, relationships, and descriptions from tables, graphs, and images.
• OCR is used to process the text in images and generate descriptions of the visual content.
• Use industry templates to organize documents such as contracts, patents, clinical trials, and financial statements.
• Structured data that can be used for searching, analysis, RAG, and agents is generated.
• Mark the items that require manual review using quality reports and self-assessments.

Axion workflow

1. Identify the business issues that need to be addressed, as well as the fields and relationships required in the end.
2. Connect to files, cloud storage, databases, or business systems, and define the scope of processing.
3. Choose an industry Schema or design a custom Operating Data Layer.
4. It performs OCR, context segmentation, table and graphic extraction, as well as entity association.
5. Check the quality reports, fields with low confidence, and abnormal documents.
6. Domain experts revise the samples, and then adjust the extraction and indexing rules.
7. Leave the task of achieving stable results to search engines, RAG, analysis tools, or Neuralith agents.

Axion supported formats and languages

Neuralith intelligent agent orchestration

Neuralith is designed for the creation, deployment, and management of enterprise-level Agentic RAG systems. It organizes search functions, models, tools, and agent workflows based on the data stored in Axion, enabling applications to answer questions and carry out tasks by leveraging an organization’s knowledge.

RSpace research workspace

RSpace brings together literature discovery, screening, thematic analysis, structured extraction, summarization, and question answering in a single workspace. Researchers can use these modules together to carry out comprehensive literature reviews, as well as to continuously monitor new papers, patents, or internal research materials.

Literature search and screening

Explore does not require users to come up with perfect Boolean search queries from the outset; instead, they start by understanding the concepts through relevant questions or seminal literature, and then expand their research accordingly. The results still need to be refined through filtering and domain expertise.

• First, clarify the research question, inclusion criteria, and exclusion criteria.
• Start concept exploration with a highly relevant paper or a complete problem.
• Examine the concept maps and relevant literature to correct ambiguous terms and missing topics.
• Combine time, database, topic, and context filtering criteria.
• Save the reasons for exclusion to avoid re-examining the same set of documents.
• Supplement the systematic review with native database searches and document the complete search strategy.

Theme analysis

Analyze identifies frequent words, concepts, and themes within a set of documents, and provides the results to the filtering panel. It is useful for uncovering the internal structure of a research collection, but the theme labels need to be interpreted by experts in the relevant field.

A high word frequency does not imply strong evidence, and thematic clustering is not equivalent to a causal relationship. Users should return to the specific paper to examine the methods, samples, statistics, and conclusions.

Structured extraction of Extract

Extract allows you to specify the fields that need to be extracted based on the Output Data Layout, and it organizes the information contained in papers, patents, or reports into comparable tables. Common fields include materials, samples, methods, indicators, experimental conditions, and conclusions.

1. Derive the minimum set of fields from the research questions, and define the type and unit for each field.
2. Select representative documents to create extraction samples.
3. Descriptions of configuration fields, synonyms, allowed values, and rules for missing values.
4. Run the extraction and conduct a random check on the correspondence between the text, tables, and graphics.
5. Standardize unit formats, entity names, and multi-value field formats.
6. Submit results with low confidence or conflicts to domain experts for verification.
7. After exporting the data, the document identifier and version are retained, allowing for rollback.

Summarize and Chat

Summarize can compress individual documents or sets of materials, while Chat allows users to ask further questions regarding the content in the workspace. Both tools are useful for quick navigation, but they cannot replace the need to read the original key texts, statistical tables, and supplementary materials.

• The abstract should distinguish between the author’s conclusions, the experimental results, and the inferences generated by the platform.
• When asking questions, specify the dataset, time frame, and meaning of terms.
• The answer is required to include references that allow one to return to the relevant document and paragraph.
• Important figures, comparisons, and negative conclusions must be verified manually.
• Do not assume that information not mentioned does not exist.

Multi-RAG and retrieval technologies

Iris.ai’s RAGaaS does not rely solely on vector similarity; instead, it combines vector retrieval, graph traversal, knowledge fingerprinting, and keyword search. The system selects the appropriate retrieval method depending on whether the question is short, long, or entity-based.

Companies can choose their own large models and use evaluation frameworks to compare different models and fine-tuning approaches. The actual accuracy depends on the interaction of data structure, retrieval methods, prompts, the model itself, and business-related evaluations.

Corporate knowledge management

Iris.ai enables the centralized collection, indexing, and enhancement of metadata for internal documents over many years, while also ensuring continuous updates, so that this knowledge can be retrieved and utilized by intelligent systems. It is suitable for industries such as research and development, manufacturing, life sciences, finance, and the public sector, where data is complex and a high level of traceability is required.

• Create fields for document owner, confidentiality level, version, and expiration date.
• Limit the visibility of data by department, project, and role.
• Set expiration rules and withdrawal rules for papers.
• Distinguish external papers and patents from internal experimental and decision-making records.
• Continuously monitor new data and perform incremental indexing.
• Ensure that each response retains the backtracking chain of documents, paragraphs, and extracted fields.

Deployment and Integration

Axion can be connected to internal databases, CRM systems, BI tools, and cloud storage; RAGaaS is also available to serve enterprise application developers. Before making a purchase, it is necessary to test the throughput, fields, Webhook functionality, authentication processes, and error recovery mechanisms using sample data, rather than relying solely on demonstrations.

Axion package and pricing

0.05 euros per page is the lowest promotional price offered by Enterprise under the best discount conditions; it is not a fixed rate that applies to all projects. The complexity of the document, OCR processing, charts, schema, method of deployment, engineering services, and level of commitment all influence the final quote.

Prices of RSpace and Neuralith

The official website does not disclose a unified subscription price for RSpace and Neuralith; quotes are provided mainly through demonstrations and assessments of the enterprise’s needs. The pages on the Internet that list prices for personal assistants with the same name as Iris are not related to this product and should not be included.

Purchasing testing methods

1. Real samples containing text, tables, scans, graphics, and multiple languages were selected.
2. Define in advance the field-level accuracy, retrieval recall, answer traceability, and processing time.
3. Have the supplier demonstrate the full processes of Axion, RSpace, or Neuralith on the same sample.
4. Check the capabilities for identifying low-confidence tags, making manual corrections, and reprocessing.
5. Separate quotes are provided for software, Credits, implementation, OCR, deployment, support, and excess usage.
6. Verify the data deletion, permission, auditing, backup, and contract termination processes.
7. Verify the metrics through small-scale production pilots before increasing the volume of data.

Privacy and external storage

The privacy policy states that Google Drive, Dropbox, and Microsoft SharePoint are accessed via OAuth2 authorization; the platform stores only encrypted tokens, not the users’ passwords. These tokens are kept in AWS environments located within the European Economic Area.

• Only grant access to the folders required by the project and to the minimum extent necessary.
• Review the sharing permissions on both the original storage platform and Iris.ai.
• When the connection is severed or the account is deleted, the corresponding OAuth tokens are revoked and removed.
• Google user data must not be used for advertising or sold, and it is subject to restrictions on its use.
• Enterprises should still determine the deletion times of document copies, indexes, extraction results, and backups.

Security and data governance

The system can reduce knowledge-related errors, but it cannot guarantee the accuracy of all extracted information and responses. Regulated decisions should still involve expert review, approval, and a designated person who is responsible for the final decision.

Account and data deletion

Users can delete their account through the account settings; the relevant data will be removed immediately and cannot be restored. If the account is deleted before the end of the paid period, no further renewal will take place, and the unused portion will not be refunded.

Accounts that have not undergone any substantial research activities for an extended period of time may be considered dormant accounts after two years, and they will be deleted following several notifications. The procedures for storing and terminating business contracts may vary; the terms specified in the signed documents shall prevail.

GitHub and the open-source status

Iris.ai has a GitHub organization that is in line with its brand identity, but currently only one repository containing the ICO contract, updated in 2018, is made public. The source code for the core platforms Axion, Neuralith, RSpace, or Multi-RAG is not available publicly.

The company states that its system incorporates both proprietary models as well as open-source models that have been modified and fine-tuned; this does not mean that the final product is open source. The open-source status in the catalog should be marked as \"no\", and the rights related to deployment and the source code must also be assessed separately.

Which users are it suitable for

• Researchers who need to discover and filter academic papers across different databases.
• A R&D team responsible for conducting systematic reviews, evidence tables, and patent analyses.
• Teams that need to extract structured fields from scans, forms, and graphics.
• AI teams in enterprises that wish to leverage internal knowledge to support RAG and agents.
• Regulated organizations that place emphasis on private deployment, permissions, auditing, and traceability.
• AI platform teams that need to compare the actual performance of models and retrieval processes in real business scenarios.

Product advantages

• It extends from findings in research papers to the structuring of multimodal data in enterprises and the orchestration of agents.
• RSpace covers Explore, Filter, Analyze, Extract, Summarize, and Chat.
• Axion can process text, scans, tables, graphics, and more than 68 languages.
• Multi-RAG combines various retrieval methods such as vectors, keywords, graph traversal, and knowledge fingerprints.
• Supports custom models, cloud, private cloud, and on-premises deployment.
• Pay attention to quality reports, low-confidence markers, evaluations, and manual review.
• Suitable for complex knowledge in fields such as science, patents, and regulated industries.

Usage restrictions

• Currently, it is mainly used for business demonstrations and customized quotes, and it is not suitable for individual users who are looking for transparent and low prices.
• RSpace and Neuralith do not disclose fixed package prices.
• Complex schemas, OCR, charts, and domain adaptation require implementation efforts as well as the input of experts.
• The extracted content, themes, and generated results may still be incorrect; it is necessary to go back to the original text to verify them.
• The technical details of the public API and the standardized information for self-service access are limited.
• GitHub does not make the code of its core products available publicly, so it is not possible to deploy a private version of the open-source software on one’s own.
• Systematic reviews still require reproducible search strategies as well as subjective decisions for inclusion and exclusion.

Frequently Asked Questions

Is Iris.ai now just a tool for searching papers?

No. The findings and research processes are focused on RSpace; the current overall scope also includes the structuring of enterprise data using Axion, as well as the orchestration of Neuralith Agentic RAG.

What are the main tools in RSpace?

It mainly includes Explore, Filter, Analyze, Extract, Summarize, and Chat; these functions can be combined to enable literature discovery, screening, thematic analysis, data extraction, summarization, and question answering.

Can it handle scanned documents and charts?

Axion supports scanning PDFs, performing OCR on images, extracting data from tables and graphics, and it can also handle Word, Excel, PowerPoint files as well as plain text.

Is Chinese supported?

Axion states that it supports more than 68 languages, including Chinese. The multilingual technical documentation still needs to be tested separately with regard to OCR accuracy, terminology usage, table formatting, and translation quality.

How much is Iris.ai?

RSpace and Neuralith do not disclose fixed prices. Axion is available in Starter, Professional, and Enterprise versions; pricing is determined through sales inquiries, with the lowest advertised price for the Enterprise version at 0.05 euros per page for large volumes.

Can it be integrated with enterprise systems?

It is possible to connect to systems such as cloud storage, databases, CRM, and BI through APIs and various connection methods; it supports deployment in the cloud, on a private cloud, or locally. The specific interfaces need to be determined as part of the project assessment.

Is Iris.ai open source?

The core products are not open source. The official GitHub account contains only an old repository for ICO contracts; the use of some open-source models in their own platforms does not mean that Axion, Neuralith, or RSpace release their source code.

Can it directly replace manual systematic reviews?

No. It can speed up the processes of discovery, screening, extraction, and summarization, but the researchers are still responsible for drafting the research protocol, assessing biases, verifying the original sources, and arriving at final conclusions.

Summary

Iris.ai has evolved from a tool for literature discovery into a knowledge infrastructure platform that covers data structuring, research processing, and enterprise agents. RSpace supports the research process, Axion converts complex data into formats usable by AI, while Neuralith is responsible for the orchestration and governance of Agentic RAG.

It is particularly suitable for organizations with complex data, a high volume of industry-specific terminology, and strict compliance requirements; however, the barriers to procurement and the complexity of implementation are higher than those associated with ordinary research assistants. Verifying accuracy, traceability, deployment, and total cost using real-world samples is key to deciding whether to adopt it.