Back to dollama.net

AI Disclosure

How generative AI was used to build dollama and this site, what AI does when you use dollama, and what the mark in the footer means.

Last updated: October 2, 2026

GENAIS level C. Content largely developed by AI but enhanced with human input or final edits. This is the site-wide level for dollama: the software, this website and its documentation.

dollama marks work made with generative AI so that you know what you are reading and running. It uses the proposed GENAIS symbols. The symbol in the footer of every page covers the whole site. Select it for a short explanation.

The mark describes how much AI was involved in producing something, not whether it is accurate or safe.

1. How dollama was made

The software. Lia Pool, the named human operator of dollama.net, decides what dollama does, specifies it, directs the work and tests the result. Most of the code is written by AI coding assistants (Anthropic's Claude models) under that direction.
This website and the documentation. Drafted by AI from the project's specifications and code, then reviewed and edited by Lia.
Evaluation results. The measurements on the Models page come from an automated test harness. The analysis is drafted by AI and checked by a person.

2. What AI does when you use dollama

Model answers are generated by AI. dollama runs open-weight language models (currently Qwen 3.5 9B and Qwen 3.5 4B) on your machines, your group's, or volunteers'. Nobody reviews an answer before you see it. Status notices and summaries marked [dollama] are written by the software, not by a model.
dollama adjusts what goes into and comes out of the model. It trims context to fit small models, repairs malformed tool calls and retries when a model stops early.
Answers can be wrong. Small models make mistakes, such as wrong answers, invented file paths and stopping before the job is done. The Models page shows how often each model succeeds without help.
Who runs what. Your tools run on your machine. The model requests file edits and shell commands, and your coding tool carries them out under its own permission prompts. dollama itself runs web searches and page fetches, and, when a request runs on a model on your own machine, reads files inside your project for sub-tasks (see the Privacy Policy §6). On the Open Network the output comes from a machine you do not control, so do not set your coding tool to approve actions automatically there.

3. The scale

The four levels, A to D, describe the split of work between a person and generative AI. The meanings are quoted from the paper.

	A	Content human developed but structure and/or format enhanced or validated by AI.
	B	Content developed by shared human and AI input. Concept developed by human but detailed by AI.
	C	Content largely developed by AI but enhanced with human input or final edits.
	D	Content entirely generated by AI prompt. May include images, video, text.

Anything on this site that differs from the site-wide level carries its own mark.

4. Reporting a problem

To report a problem with dollama or this site, including a security issue, see Security reports.

5. Source

Glassey, S. (2025). Standardising AI Disclosure in Academia: The GENAIS Symbology Approach. Preprints. doi:10.20944/preprints202510.2183.v1. The paper is CC BY 4.0; the symbols are reproduced from it with attribution. GENAIS™ is the author's trademark.