Assistant versions

AI tools have a habit of changing overnight. ChatHQ takes the opposite approach: nothing your visitors talk to changes until you publish it, and every published state is kept as a numbered version you can go back to.

Draft and live

Each assistant has up to two versions in play at any time:

Live — The version your website visitors and API integrations are talking to right now.
Draft — Your working copy. Changes to instructions, knowledge sources, model or settings go into the draft first.

Editing never touches the live version. You can take as long as you need to adjust instructions, add documents and test in the test chat. Visitors keep getting the version you last published.

Publishing

When the draft is ready, publish it. The draft becomes the new live version and receives the next number (V1, V2, V3 and so on). The previous live version is kept in the history.

If there is nothing new to publish, the assistant shows No unpublished changes.

Restoring an earlier version

If a change made answers worse, you do not have to undo it by hand. Pick an earlier version from the history and restore it. The restored configuration becomes your draft, marked as restored from that version, and you publish it like any other change.

Note:

Restoring brings back the assistant's configuration: instructions, model, settings and the set of knowledge sources it had at the time. Conversation history is not affected.

Version history

The history shows, for each version, when it was edited or published and a summary of what changed, for example configuration updated or +2 files added. It is the quickest way to answer "what did we change before the answers got worse?".

The engine behind it

The same discipline applies to the AI engine that powers every assistant. Each major change to the engine is released as a numbered version, tested against the previous one, and can be rolled back if answer quality drops. See Release testing.

Was this article helpful?