Since the AI bubble is inexorably upon us, it's time to talk about it.
AI advances have boosted productivity in Thunderbird/Betterbird development. In the Thunderbird project, various folks have solved bugs with AI-assisted coding tools, powered by models such as OpenAI's GPT family, Anthropic's Claude models, Google's Gemini models, and DeepSeek models. They have analysed code, proposed and reviewed patches, written tests, and have even run the entire development pipeline via agents right up to running new tests and submitting work to Mozilla's Phabricator platform.
Our project hasn't spend big on AI subscriptions, but with humble subscription to DeepSeek v4 Pro, we've been able to analyse issues (see articles's screenshot), written scripts and automated tedious tasks like localisation, since one new string needs to be added to 15 localisation files and also be made consistent with existing translations in Thunderbird. So far, we haven't produced a Betterbird patch with AI yet, but we're looking forward to the first adequate task.
Update: We had DeepSeek revise our patch for showing native alerts on the application screen. AI pointed out some shortcomings wrt. to screen scaling and added code to cater for monitors aligned vertically. As a reminder, the function can be enabled with preference mail.notification.show_on_app_screen.
But AI is not only helping developers, users also benefit by having AI summarise messages or help when writing message by rephrasing, summarising, explaining, translating, etc. Here's a table of available add-ons.
Disclaimer: We've only surveyed listings on ATN and not tested those add-ons. There is no relationship between us and any of the add-on authors. If you are an add-on author and we misrepresented your add-on, please get in touch.
| Add-on | Users | Comment |
|---|---|---|
| AI Anywhere for ChatGPT | 4467 | Fixed set of models, apparently no local AI |
| AI Mail Support | 783 | Various models, incl. via Ollama, multiple prompts |
| ThunderAI | 13422 | Various models, incl. via Ollama, multiple prompts |
| AI Email Assistant | 323 | (few details, apparently OpenAI API support) |
| AI Mate | 122 | (few details, apparently OpenAI API support and variable prompt) |
| Mail2Cal-AI | 60 | For Calendar, Ollama only |
| AI Mail Extended | 62 | (few details, apparently local AI) |
| スマート返信AI | 171 | (Smart Reply AI, in Japanese) |
| GENA AI-Powered Email Assistant | (unknown) | Various models, incl. via Ollama, auto-tagging, thread summary, semantic search |
| ChatGPT Quick Access | 1275 | (ChatGPT tab) |
| ChatGPT Addon | 345 | (few details, apparently ChatGPT only) |
For e-mail processing, it makes sense to use local models, for example via Ollama. We also gave LM Studio a try, but found the model download unbearably slow on a fibre optic connection. Using a local model, e-mail data is not sent to the AI provider's data centre for processing.
Ollama users should note that in order for their add-on to be able to contact the local server at http://localhost:11434, they need to work around a CORS (Cross-Origin Resource Sharing) issue leading to a 403 error. Ollama needs to send the Access-Control-Allow-Origin: * header which can be enabled by setting the environment variable OLLAMA_ORIGINS to a value of *.







(Screenshot taken using some newsgroup content.)

