Startup Geniez is stepping out of stealth mode, developing software to connect AI Large Language Models and agents with mainframes and extract real-time data for AI use.
LLMs and agents typically query x86 server and unstructured data held on storage arrays or in the public cloud. They use existing search methodology to look at structured data in transactional databases, as well as knowledge graphs; ones from Neo4j for example. But the enormous amounts of data, both structured and unstructured, stored in IBM’s walled mainframe garden has been left unseen except by IBM’s own watsonx AI facilities with its Data Gate for watsonx.
Geniez is opening up the mainframe’s data storage for real-time use by non-IBM LLMs and agents.
The company was founded by CEO Gil Peleg and Dan Shprung who were both involved with Model9, an Israel-based startup bought by BMC in April 2023. The two became part of BMC’s AMI organization.
Peleg told B&F: “We’re going to bring GenAI to the mainframe just like in Model9 we brought object storage to the mainframe.”
He added: “We gained a lot of experience at Model9 when we started. … [But] we didn’t realise the powers, the dynamics in the market, the huge customers, the huge incumbents, how deeply customers are reliant on IBM and how to get them to change.”
Mainframe customers, all 20,000 to 25,000 of them, are heavily reliant on their machines, their systems of record, and migration off mainframes is not as frequent as some may assume. The data on these systems is critical for the banks, insurance companies, credit card suppliers, airlines, shipping companies, retailers and manufacturers that have mainframes. These enterprises are all aware of AI and understand that the more of their data which can be accessed by the LLMs and agents they use the better will be the responses they get.
But many won’t move the data off the mainframes to achieve that. So it has to be fetched. But by what? There are myriad mainframe data sources with individual database protocols, record formats, and access methods. Establishing any external access requires code, specific to a target data source to be written, tested and executed on the mainframe. This provision of point-to-specific-point extract, transform and load (ETL) procedures can take months of effort.
There are existing ETL procedures to get mainframedata moved into data lakes, but the data is not real-time. What Geniez claims to be offering is a ready-made collection of real-time, data extract connectors – a mainframe databot so to speak. Peleg says it provides real-time direct access to mainframe data sources, with support for mainframe data sources including: DB2, IMS, MQ, VSAM, and Data sets. Supported LLMs and agents include ones from Meta (llama), Anthropic, OpenAI, Gemini and Amazon Bedrock.
Geniez claims its software has mainframe-grade reliability, availability and security, RACF controls, and end-to-end encryption. There are debugging, observability, governance and audit capabilities, plus a standard SDK for python applications running anywhere.
Geniez has a demonstrable prototype product and, Peleg said: ”We have already closed a seed funding round from top US investors as well as some of our previous investors at Model9.”
