ArgueSecure 2.6.0
Two better models for the optional AI assistant, installed with one command, plus standards templates and a rewritten help centre.
The assistant
Before this release the assistant shipped with a small default model. Tested on the kind of machine that runs it (four CPU cores), asked why a total does not change when a reduction target is set, the default answered correctly once in four tries and invented figures in one of four drafted summaries. Two new presets answered correctly four times in four and invented no figures:
- small (Gemma 4 E4B) for modest hardware.
- large (Gemma 4 26B-A4B) for a machine with more memory.
Neither preset is easier or harder to steer than the old default, and nothing a model writes reaches an assessment until a person accepts it.
- One command installs a model:
docker compose run --rm assistant-install small(orlarge). It checks free memory and disk, resumes interrupted downloads, and verifies the file before using it. A model of your own still works. - The admin AI tab shows what is really installed, asked of the running service, rather than a label typed into a settings file, and lists both presets with the memory and download each needs.
- Faster after a restart. The help answers are prepared in the background when the service starts, so the first person to ask does not wait a minute or two.
- Better defaults. Four processing threads instead of eight, and the service’s CPU limit follows that setting.
Also in this release
- Standards templates for ISO/IEC 27001, ISO/IEC 42001, the EU AI Act, SOC 2, GDPR, NIS2 and the CIS Controls, bringing the bundled templates to twelve. Each control is a defence you can mark as implemented, planned or shelved, which gives a gap analysis against the standard.
- Add from template on any item, and From template in the new-item form, copy individual risks, attacks or defences into an existing assessment.
- A help centre rewritten for the current product: ten pages that also feed the assistant’s answers.
- Your own edits no longer come back to bite you. Your own change used to echo back from the live connection and could overwrite what you had just typed, or blank the Details tab on a new item.
Upgrading from 2.5.x
- A 2.5 assistant keeps running without changes. Whatever model your settings named, or the old default, still resolves to the same file.
ASSISTANT_THREADSnow defaults to 4. A machine with fewer than four CPUs to give the assistant must lower it or the service will not start.- To move onto a preset: remove
ASSISTANT_MODEL_FILEandASSISTANT_MODELfrom.env, run the install command above, thendocker compose --profile assistant restart assistant. The old model file can then be deleted. /assistant/statusno longer reports a model name. Read it from the admin AI tab instead.- Templates added in this release load for new installs; see the README to load them into an existing instance.