Tessovis

Tessovis is an ontology runtime for enterprises that keeps its data on its own servers. It takes the records scattered across ERP, MES, WMS, databases, files, and event streams and maps them to business objects with meaning, relationships, permissions, and actions. People and AI query, decide, and write back on that one graph, and every step can be audited. A free Community Edition ships with readable source. I have led it as CEO since February 2026.

The kernel turns tables, fields, and APIs into five things: Object, Property, Link, Function, and Action. A factory, a work order, a material, a supplier, a stock level, and a person become objects that carry their own rules. Data does not have to move first: a virtual table joins a source system into the graph, and moved data keeps its lineage. Every write goes through an Action transaction that validates, audits, and writes back to the source system. There is no back door to the database.

AI works as an operator with permissions. An agent can do what the person who started it can do, cut down to a whitelist. Actions are staged by default and take effect only after approval; automatic release is granted explicitly and can be withdrawn within ten seconds. The reference flow on the site is a stock monitor that fires when a part drops below its safety level, an agent that stages a restock order with its parameters and evidence in view, a planner who approves or rejects, and a write-back that adds one record to the audit stream.

Around the kernel sit data access, AI orchestration, applications and analytics, and OSDK. Dashboards, simulation, and process mining all read the same graph. A simulation runs in a Simulated World beside the Observed World, so a rescheduling or inventory experiment edits a sandbox and never the main line. OSDK generates a typed SDK for the ontology in about five minutes. Two reference ontologies come in the box: manufacturing and supply chain, with work orders, materials, inventory, suppliers, and scheduling, and a general operations template for assets, processes, and approvals.

It installs in your own machine room. The acceptance criterion for a single node is thirty minutes or less from bare metal to the login page, on a fresh Ubuntu 24.04 LTS with image pulls included. The offline package makes zero outbound connections in its factory configuration, which a packet capture can confirm. Tiers grow from one node to three to a Kubernetes cluster on the same ontology, and the Enterprise Edition's M and L tiers carry a 99.9% annual availability target.

The Community Edition runs every module on one node, offline, with readable source. The Enterprise Edition adds high availability, SAML, dual-person review, and optional Chinese national cryptography. A hosted edition and a six-to-ten-week implementation package exist for teams that want the work done. The entry path is a two-week PoC workshop on the customer's data in the customer's machine room, ending with one working chain from data access to an AI-staged, human-approved action.