Version 2026.10.04Released:
Agentique 2026.10.04, the on-premise edition
The whole application on the organisation’s own server, in Docker containers, with no external services and no internet connection needed.
- Local AI on the server’s processor: open-source language models (Qwen3.5 and Qwen3.6, Apache 2.0 licence, through llama.cpp), meaning-based search and the reading of scanned documents.
- The assistant works with every tool of the application that needs no internet; deletions, data changes and e-mails wait for a person’s confirmation.
- The language model is chosen and downloaded in the settings: each download is checked against its published SHA-256 sum, and if a new model does not start, the previous one comes back by itself.
- Three levels of effort, Quick, Normal and In depth, for questions and flows.
- Security for private installations: HTTPS enforced by the server, two-step sign-in, separate internal networks for the AI and the data and, optionally, access limited to chosen networks and devices.
- Updates with a verified backup first and an automatic return to the previous version if the checks after the update fail.
- An extended activity log, and the organisation’s memory, used in answers automatically.