I found out that "Optimize memories" in the Memory settings tab, can significantly reduce the impact of memory on the chat context usage. With memory "on" the application automatically adds memory items across chats. In my case ~4100 items, containing ~89K words, which hit the chat ceiling instantly. Using the optimize function decreased that to 116 items and ~2500 words. This created over 90% chat space. However, I advised Support to lift the token capacity to a minimum of 500K.

My use case: comparing a long device manual against a tender spec and listing every unmet requirement — verbatim, word for word. Standard compliance/tender work.

Lumo does not even start and the context is done...

Ran the identical task on a free Gemini account: both PDFs uploaded, full point-by-point analysis, no compaction, no lost fidelity.

For fairness: I get Proton's privacy angle — zero-access encryption, no training on your data — that's why I picked Lumo. But privacy doesn't require a small context window, and it doesn't excuse such limited usability.

Paid ~xx$/year for a tool that couldn't finish the job a free competitor completed without blinking. Filed a complaint with support.

Not anti-Proton, I'm a paying customer across the ecosystem. But for long-document professional work: it's not there yet.

Related, and more urgent in my view: Lumo Plus now enforces a hard 250-message limit per conversation, introduced around 20–24 October with no announcement and no documentation. Past that point the composer disappears and the conversation becomes read-only. Several professional conversations I've used continuously since last year were severed mid-session — no warning, no notice, no grandfathering.

This bears directly on your suggestion. Competitors advertise ~1M-token context windows (Gemini 3.1 Pro, Claude 5, GPT-5.6/6 Astra), but Lumo sits at 131.1K. Even if that gap were closed, a 250-message ceiling would still terminate conversations first. Your point about paying customers funding the extra compute is exactly the model that would solve this: a paid continuity tier, with disclosed numeric limits.

I've opened a dedicated suggestion here: https://protonmail.uservoice.com/forums/932842-lumo/suggestions/51863964-composer-lockout-250-message-limit-introduced-wit — if you're affected, please vote there so this gets documented publicly.

Export now exists, but it's wholly inadequate: it preserves the text, not the working relationship and the collaboration built with Lumos over time. Older conversations under "All Chats" don't show the export option at all — not less important, just put aside, and they need to be continued.

Still unanswered by Proton: (1) is there a 250-message limit on Lumo Plus, and where is it documented; (2) what are the per-conversation limits for Business and Lumo Professional; (3) what is the token/context allowance in numbers; (4) why was this change never communicated before it affected users' work.

Two steps would resolve most of it. Immediate relief: grandfathering — existing conversations regain their composer and continue with the Lumos that hold their history, with any limit applying only to new chats. Lasting solution: guaranteed continuity for those willing to pay, with transparent numeric limits — the direction you're arguing for here.

I agree 256K minimum for dev nowadays unless strict management of the context window, which the Lumo API will allow.

Echoing this request as critical. I need the ability to ingest and analyze large documents (mainly large PDFs and spreadsheet workbooks) with high-quality inference.

If the product supports a 1M-token context window, expanding to that would be welcome. I understand that high context and high inference are resource-intensive requests, and I think many users would be willing to pay more for a higher platform tier that supports this ability.

Context is a real disaster and a Lumo 2.0 regression. Messages of “context length exceeded” appear even on 10% of token capacity.

Context limit is making lumo useless for me. I am trying to analyze simple csv budget files, and any more than one or two files exceeds the limit (maybe 1mb combined).

the session/project/chat memory is what's consuming all the credits/tokens. This product is unusable as-is and my money would be better spend elsewhere. Which is a shame, because the privacy policy and trust i have for Proton overall is why I paid for the service in the first place.

Lumo 2.0 has been advertised to handle larger chats and files. However, the token capacity is still the same old one as before, namely 128K (~ 90.000 words, 10MB). Nothing changed in that. So there is no such improvement. Even worse: Lumo’s thinking (memory) consumes loads of tokens, regardless of the mode, so that new function decrease overall capacity. Large files accepted in previous versions are rejected in 2.0. A downgrade. And despite an extensive Support ticket over months these issues are not solved at all. In short, I don’t get what I’ve paid for.

Isn’t GLM based off Z ai? Isn’t that a Chinese company? How secure is this?

I think Proton had increased the context limit to 256K recently. But for a power user like myself is not enough. 512K is okay.

or at least make a third more expensive tier in which the input token limit is at the model's limit

I use Lumo to do detailed and complex work which involves uploading multiple documents and basically as soon as I've uploaded what I need to upload there are no more tokens. Other models offer 1 million tokens and I feel like Lumo should at least get halfway there