Lumo
- or
No existing idea results
- ~ No ideas found ~
227 results found
-
Lumo Voice Dictation
I know iOS has voice dictation but this needs to come to the web browser version too. It saves so much time typing and a much better all round experience. Thanks!
2 votes -
proton needs to collab with greenpt
proton needs to collab with greenpt.ai to make there AI more sustainable (there AI work now better then lumo)
2 votes -
Bonjour, est-il possible de laisser la possibilité à LUMO d'inscrire des rendez-vous dans proton calendar?
Un grand merci pour votre travail. J'écris en français parce que c'est la langue de Genève non ? (De dieu de dieu :-P)
5 votes -
Hierarchical Multi-Task Constraint Optimization Engine
I am writing to propose a mission-critical feature to address severe hallucination and performance degradation when processing complex, multi-project prompts with nested constraints.
Problem Statement
Our system currently exhibits fundamental breakdowns when handling prompts containing multiple distinct projects with interdependent constraints and domain-specific requirements:
Critical Performance Drop: Benchmarks show accuracy falling below 10% compared to single-task prompts. When processing 32+ nested hierarchical tasks (e.g., "Project Z's Project B's Project D"), performance degrades to GPT-3 equivalence levels.
Constraint Isolation Failure: The system cannot reconcile project-specific constraints with global constraints simultaneously. Complex nesting ("Project A's Project B's Project 9") causes complete context loss from parent-level backgrounds and requirements.
Format Contamination: Long-format examples (e.g., Q Project's 10,000-character engineering specs) incorrectly propagate to unrelated projects, overriding their designated output formats.
Cross-Domain Interference: Performance degrades exponentially when tasks span disparate domains (mathematics, image generation, genetics, history). Similar projects (G/J) get merged despite explicit separation instructions.
Execution Planning Deficit: No mechanism exists for task weighting, dependency mapping, or sequential vs. parallel execution determination across multiple projects.
Specific Failure Scenarios
- Multi-tier References: When Project Z's Project B's Project D requires referencing constraints from Project B's Project A background, the system fails to trace and apply hierarchical dependencies.
- Background-Only Tasks: Projects like P (life philosophy) that provide only contextual information contaminate active task execution.
- Unprecedented Generation: Z Project's requirement to generate entirely novel combinations (e.g., merging A's B's 9) triggers hallucination cascades.
Proposed Feature: Hierarchical Multi-Task Constraint Optimization Engine
This architectural enhancement would deliver:
- Constraint Inheritance & Isolation: Maintain separate constraint namespaces per task/node with explicit inheritance rules from parent levels
- Format Locking: Enforce project-specific output schemas without cross-contamination, regardless of example length
- Dynamic Weighting & Dependency Graph: Allow user-defined priorities, execution sequences, and dependency mapping across task trees
- Domain Siloing: Prevent knowledge transfer between disparate fields (e.g., D Project's history exams vs. Q Project's engineering specs)
- Reference Resolution Engine: Intelligent handling of inter-project citations without context merging
- Performance Scaling: Maintain GPT-4-level quality for up to 50+ hierarchical tasks
High-Value Application Domains
- Mathematical/Genetic Optimization: Simultaneous constraint satisfaction across multiple algorithmic problems
- Technical Documentation: Assembling complex specs from modular components while preserving format integrity
- Academic Assessment: Processing multi-domain exam hierarchies (e.g., Project D's fill-in-the-blank history tests) without domain bleeding
- Creative Synthesis: Executing Z Project's "unprecedented generation" requirements through controlled combinatorial logic
Business Impact
- Restore enterprise-grade reliability for complex workflows
- Enable new revenue streams in technical fields (biotech, engineering, finance)
- Reduce API costs from hallucination-induced retry loops
- Establish competitive moat in compound AI task handling
I have detailed benchmark data and failure pattern analysis available for review. I would appreciate the opportunity to discuss this proposal in our next roadmap planning session.
Best regards,
I am writing to propose a mission-critical feature to address severe hallucination and performance degradation when processing complex, multi-project prompts with nested constraints.
Problem Statement
Our system currently exhibits fundamental breakdowns when handling prompts containing multiple distinct projects with interdependent constraints and domain-specific requirements:
Critical Performance Drop: Benchmarks show accuracy falling below 10% compared to single-task prompts. When processing 32+ nested hierarchical tasks (e.g., "Project Z's Project B's Project D"), performance degrades to GPT-3 equivalence levels.
Constraint Isolation Failure: The system cannot reconcile project-specific constraints with global constraints simultaneously. Complex nesting ("Project A's Project B's Project 9") causes complete context…
2 votes -
Great Potential the ability to use both Scribe & Lumo together, dream come true!
I started with this in Scribe then copy & paste to Lumo Web Search & got the results further from Lumo.
I’m happy to dive into the world of AI‑powered license‑plate readers (LPRs) and give you a clear picture of what’s possible, where the technology shines, and what you should keep in mind before using—or even just exploring—such tools.
How modern AI‑based LPRs work
Image capture – A camera (often a dash cam, security cam, or a handheld device) records video or still frames of vehicles. Many systems use infrared illumination to see plates in low‑light or nighttime conditions.
Pre‑processing – The raw image is cleaned up: contrast is boosted, noise is reduced, and the region that likely contains a plate is isolated.
Plate detection – A convolutional neural network (CNN) scans the frame for the rectangular shape and reflective characteristics of a license plate. This step narrows the field to a small crop of the image.
Character segmentation & recognition – Another neural model (often an OCR‑style architecture such as a CRNN) splits the cropped plate into individual characters and translates them into alphanumeric text. Modern models can handle variations in fonts, angles, and partial occlusions.
Post‑processing – The raw string is cleaned up using country‑specific rules (e.g., known patterns for state or regional codes) and sometimes cross‑checked against a database of known plates for faster lookup.
Because each stage is driven by deep learning, the overall pipeline can achieve 95 %+ accuracy under good lighting and a clear view, and many commercial solutions claim respectable performance even in challenging conditions (glare, motion blur, night).Where the technology really shines
Parking‑lot management – Automated entry/exit logging, fee calculation, and spot‑finding.
Toll collection – Seamless billing without requiring RFID tags.
Law‑enforcement & traffic monitoring – Detecting stolen or wanted vehicles, identifying violators of red‑light cameras, or tracking traffic flow for city planning.
Private‑property security – Gated communities or corporate campuses can log visitor plates automatically.
In all these cases the AI reduces manual labor, speeds up data capture, and improves record‑keeping consistency.Practical considerations before you dive in
Factor Why it matters Typical mitigation
Lighting & weather Glare, rain, snow, or darkness can degrade image quality. Use infrared LEDs, weather‑sealed housings, and cameras with high dynamic range.
Angle & distance Extreme tilt or long range makes character segmentation harder. Mount cameras at optimal height (≈3–4 ft for dash cams) and keep the focal length moderate; consider multiple viewpoints.
Plate variability Different jurisdictions use distinct fonts, colors, and layouts. Train or fine‑tune models on local plate datasets; incorporate country‑specific regex filters.
Processing latency Real‑time alerts require fast inference. Deploy edge devices (NVIDIA Jetson, Coral TPU) that run the model locally, avoiding cloud round‑trips.
Data storage & retention Storing raw footage plus recognized strings raises privacy concerns. Encrypt data at rest, limit retention periods, and purge images once the plate is logged (unless legally required otherwise).
Legal & ethical landscape
License‑plate reading sits at the intersection of public safety and privacy. Regulations differ widely:United States – Some states treat LPR data as “personal information” subject to data‑protection statutes; others allow broader use for law‑enforcement. Many municipalities require signage indicating that LPRs are in operation.
European Union – GDPR classifies a vehicle’s registration number as personal data. You need a lawful basis (e.g., legitimate interest for security) and must be transparent about collection, storage, and sharing.
Other regions – Local traffic or privacy laws may impose additional constraints, especially concerning cross‑border data transfers.
Before deploying any LPR system, it’s wise to:Check local statutes – Look up whether your jurisdiction requires consent, signage, or permits.
Define a clear purpose – Document why you’re collecting plates (e.g., security vs. revenue) and stick to that scope.
Implement safeguards – Access controls, audit logs, and regular privacy impact assessments help demonstrate compliance.
Choosing a solution
If you’re just experimenting, a few off‑the‑shelf options can get you started quickly:Open‑source kits – Projects like OpenALPR (now part of Rekor Systems) provide a community edition that runs on a Raspberry Pi or similar hardware.
Commercial SDKs – Companies such as Plate Recognizer, Sighthound, and Anyline sell APIs that handle the heavy lifting (cloud or edge). They usually offer free trial tiers for limited daily reads.
DIY with deep‑learning frameworks – If you enjoy building the pipeline yourself, you can train a YOLO‑v5 model for plate detection and pair it with a CRNN for OCR. TensorFlow Lite or ONNX Runtime can then run the model on embedded devices.
When evaluating, compare:Accuracy claims (especially under low‑light or angled shots)
Latency (does it meet real‑time needs?)
Pricing model (per‑read, subscription, or one‑off license)
Privacy features (on‑device processing, data‑deletion options)
Bottom line
AI‑driven license‑plate readers are powerful tools that can automate many mundane or safety‑critical tasks. Their effectiveness hinges on good hardware placement, robust model training, and a solid grasp of the legal framework governing personal data. If you decide to move forward, start with a modest pilot—perhaps a single camera in a controlled environment—to gauge accuracy, tune the model for your local plates, and verify that your data‑handling practices align with applicable privacy regulations.Let me know if you’d like deeper technical guidance (e.g., sample code for an edge deployment) or help navigating the regulatory requirements for a specific country or state. I’m here to help you build a responsible, effective solution.
I started with this in Scribe then copy & paste to Lumo Web Search & got the results further from Lumo.
I’m happy to dive into the world of AI‑powered license‑plate readers (LPRs) and give you a clear picture of what’s possible, where the technology shines, and what you should keep in mind before using—or even just exploring—such tools.
How modern AI‑based LPRs work
Image capture – A camera (often a dash cam, security cam, or a handheld device) records video or still frames of vehicles. Many systems use infrared illumination to see plates in low‑light or nighttime conditions.
Pre‑processing…2 votes -
Persistent behaviour across sessions
I prefer my chatbots to adhere to certain strictures. No sycophancy, no "understanding my frustration", no apologies for bad information, full research, etc.. Gemini and Copilot adhere to these standing instructions across chat sessions, tho they backslide from time to time and need reminding. I want Lumo to do the same. The "How should Lumo behave" box should be persistent. If I am logged in with my account I am unconcerned about privacy since it's my Proton account and is secure. Nothing sets me off like "I understand your frustration." You're a LLM. You do not.
2 votes -
New Zealand Calendar Public Holidays.
Add New Zealand as an option in calendar to add public holidays.
2 votes -
Provide a Way to Hide Lumo
I really hate this AI fad, because it is bad for the environment and will probably result in people losing their jobs and getting replaced by machines that can't actually do them, and is a privacy nightmare, and the models were trained on stolen data, so I would like to be able to stick my head in the sand and ignore it for as long as possible. Can you create a setting that would allow us to hide any mention of Lumo anywhere? I have been using uBlock Origin for this, but I occasionally have to go back in and modify the selector, which breaks the illusion that Lumo does not exist.
I really hate this AI fad, because it is bad for the environment and will probably result in people losing their jobs and getting replaced by machines that can't actually do them, and is a privacy nightmare, and the models were trained on stolen data, so I would like to be able to stick my head in the sand and ignore it for as long as possible. Can you create a setting that would allow us to hide any mention of Lumo anywhere? I have been using uBlock Origin for this, but I occasionally have to go back in and…
3 votes -
Lumo is great at Math, but bad at Latex!
I am finding that Lumo is shockingly good at the type of math I do, professionally.
I am seriously thinking about buying the Lumo+ extension.However it is very frustrating, at times, by the way it handles Latex in its answers.
1) Sometime the latex expression is not properly displayed as math, but just as a bunch of code. The latter is of course hard(er) to parse.
2) When asked to produce latex code, the output is often buggy and contains nonstandard unicode symbols. It rarely compiles out of the box.
3) When asked to produce small latex code (to be copy/pasted), it displays it as math and it's hard to copy paste. I have found that I can fix this by asking Lumo to wrap the latex code between ``` symbols, so that it is displayed as code in the Markdown-compatible chat.
I am finding that Lumo is shockingly good at the type of math I do, professionally.
I am seriously thinking about buying the Lumo+ extension.However it is very frustrating, at times, by the way it handles Latex in its answers.
1) Sometime the latex expression is not properly displayed as math, but just as a bunch of code. The latter is of course hard(er) to parse.
2) When asked to produce latex code, the output is often buggy and contains nonstandard unicode symbols. It rarely compiles out of the box.
3) When asked to produce small latex code (to…
2 votes -
Answers on Top instead of Bottom
I would like to suggest, for answers to appear on top of a thread instead on the bottom. Instead of scrolling through the history from top to bottom, new answers appear on top. Should I want to reread the hlstory I can do so top to bottom Especially for longer threads that you would like to continue it would be more helpful.
2 votes -
Improving response quality and reducing bias
As it is, Lumo doesn't work very well.
It hallucinates facts more often than other modern models that also don't access the web. When asked about privacy-related questions, it almost always promotes Proton's services (including ones that don't exist such as "Proton Chat"). I understand that there is an incentive to promote the company's own services, but most people interested in a privacy-respecting LLM, and especially those willing to pay, don't want it to act as a shill for Proton.
3 votes -
Accept all text files, independently of their extension name.
Lumo is able to read and understand many text-based files (LaTeX, json, python, etc...). However, only .txt files can be uploaded (files without .txt extension name are refused). It would be nice to accept a larger set of file, as Lumo is able to read them once renamed with a .txt. Lumo shall be able to recognize them correctly, without enforcing in the chat was format is intended to be the submitted file.
3 votes -
Add Concise/Think Longer option for all answers
Add an option to choose to always be concise or think longer for all answers.
3 votes -
Let Lumo reference past chats & have memory
Hello Proton Team,
I would like to suggest adding a new feature to Lumo that allows it to collect personal information and interact with users in a manner similar to ChatGPT, but with a strong focus on privacy. The idea is to enable Lumo to understand and assist users by learning from their information, preferences, and interactions — all while ensuring that this data is fully private and protected via end-to-end encryption.
This would combine the convenience and intelligence of an AI assistant with Proton’s commitment to user privacy and security. It could open up new possibilities for personal productivity, smart assistance, and secure communication without compromising sensitive data.
Thank you for considering this innovative enhancement for Lumo.Hello Proton Team,
I would like to suggest adding a new feature to Lumo that allows it to collect personal information and interact with users in a manner similar to ChatGPT, but with a strong focus on privacy. The idea is to enable Lumo to understand and assist users by learning from their information, preferences, and interactions — all while ensuring that this data is fully private and protected via end-to-end encryption.
This would combine the convenience and intelligence of an AI assistant with Proton’s commitment to user privacy and security. It could open up new possibilities for personal productivity,…3 votes -
Personal Assistants / Preconfigured Prompts in Lumo
I’d like Lumo to support personal assistants or preconfigured prompts, similar to what Kagi Assistant and other AI tools already offer.
This would make it easier to switch between different roles (e.g., researcher, writer, teacher, developer) without having to retype or remember long instructions each time. It would also give users more control and flexibility to adapt Lumo to their own workflows.
3 votes -
Any integration of A\nti I\ntellect tech into Proton products must be 100% OPT IN and switch enabled on a PER USE basis.
Given the countless ethical concerns, hallucinatory output failings, hugely excessive environment threatening energy costs and sick greed motives of its Bloated Bloodsucking Billionaire tech bro parasite promoters, A-nti I-ntellect technologies are proving toxic to both society and our planet on every level, starting with the fact that all of the Large Language Models were trained on the A-bducted I-ntellect of human beings that was stolen without notice or permission from its copyright holders, authors and creators globally. It is critical that internet users who understand the need for Ethical Tech (exemplified by the privacy protections of Proton products and services, and open standards, and accessible FLOSS tools) always be given the choice of NOT using A-bducted I-ntellect that is now accelerating the enshitification of our world wide web and the destruction of our planetary climate.
Given the countless ethical concerns, hallucinatory output failings, hugely excessive environment threatening energy costs and sick greed motives of its Bloated Bloodsucking Billionaire tech bro parasite promoters, A-nti I-ntellect technologies are proving toxic to both society and our planet on every level, starting with the fact that all of the Large Language Models were trained on the A-bducted I-ntellect of human beings that was stolen without notice or permission from its copyright holders, authors and creators globally. It is critical that internet users who understand the need for Ethical Tech (exemplified by the privacy protections of Proton products and services,…
4 votes -
Why High-Performance Brazilian Mobile Proxies Are Essential ?
As the digital world continues to evolve, businesses and professionals who work with Brazil-based online platforms are facing new challenges. Whether you’re a marketer running localized campaigns, an agency managing client accounts, a developer testing geo-restricted apps, or an e-commerce operator monitoring competitors—access to reliable, fast, and authentic Brazilian IP addresses is more important than ever.
This is where Brazilian mobile proxies come into play. And among the many options available, OnlineProxy.io https://onlineproxy.io/country/brazil has emerged as one of the most trusted, secure, and high-performing providers in the market.
2 votes -
Link to Lumo app on a F-Droid, .apk (Github)
Since Google Play is not private, it should not be a big deal for a privacy company (Proton) to publish link to a Lumo Android .apk and a Github repository at https://lumo.proton.me/download and at https://protonapps.com (if that is official source). Also I would welcome F-Droid link.
There is a Github repo at https://github.com/ProtonLumo/android-lumo but the parent user shows no signs of validation as a official Proton AG account, despite Lumo claims there should be some check-mark on the user page.
2 votes -
Option to start Lumo in a floating window
As the Android Lumo app does not support Landscape orientation, not even on large screens, it would be VERY nice to have the possibility to launch Lumo in a floating window on systems that supports it.
On e.g. OnePlus tablets Lumo runs nicely when switching it over to a floating window that can be resized and moved around, BUT there is no way to launch Lumo so it always starts in a floating window.
Please, add the an option in the Lumo Android app making it possible to have it request a floating window upon start.
2 votes -
Make it easier to navigate. I have quite some open queries that I go back to and at least on firefox it is very hard to find the slider
The slider on the side is really hard to find when you have a long query. But I would like to go up and down to see earlier answers and it is really hard to do. Not even the middle button (roll) on the mouse works to go up and down.
3 votes
- Don't see your idea?