#75 [feature] group chat message summary

[Anonymous]

Originally created by: supreme-gg-gg
Originally owned by: endernoke

Is your feature request related to a problem? Please describe.
Most productivity (commercial) messenger apps like Google Chat Groups have features that summarise group / thread discussion with AI so that you don't have to read over 20+ messages.

Somehow meta supports great AI at making meme stickers but not messages summary, great priority choice!

Many modern group chats of young people easily pile up of 100+ messages most of which are complete irrelevant useless garbage and if you end up scrolling through them you either waste your time or miss the 1% important message in between brianrot. This is where we make Instagram great again

Describe the solution you'd like
We plan to integrate Gemma3n, a fully open sourced LLM by Google DeepMind that is most performant while lightweight model available, to summarise group chat messages FULLY PRIVATE AND LOCALLY. There will NOT, I repeat, NOT be anything sent to dark corners of the Internet -- we value your privacy. This model is so lightweight that it's designed for edge devices and smartphones on-device inference that most people's laptop (unless you're using Windows XP in your lab) would be capable to download and run locally (with only a CPU). Of course, this feature would not be enabled at all automatically in the future release. You need to toggle this explicitly.

Describe alternatives you've considered
Send your private messages to ChatGPT and let OpenAI steal your information 😂

Additional context
Add any other context or screenshots about the feature request here.

[Anonymous]

Originally posted by: supreme-gg-gg

Look into Gemma3n or Qwen try to find lightweight models, perhaps also have an inference API option

[Anonymous]

Originally posted by: supreme-gg-gg

I think this would be great for first time contributors because this does not involve any in depth understanding of codebase logic like how UI is rendered, this is a pretty much standalone feature (even though its scope might be big)

[Anonymous]

Originally posted by: supreme-gg-gg

Update: would be better if we let the user setup a inference server like Ollama or vLLM locally and then the instagram CLI just interacts with that for all LLM features like this one, this keeps everything very lightweight and scalable

[Anonymous]

• assigned_to: endernoke

[Anonymous]

Ticket changed by: endernoke

• status: open --> closed

[Anonymous]

Originally posted by: endernoke

Closed by [#118] [#132]