Lumo
- or
No existing idea results
- ~ No ideas found ~
227 results found
-
Lumo Client (or API) with secure local execution of Lumo‑generated Python graphical output
When users request visualizations, Lumo frequently replies with Python code snippets (e.g., matplotlib graphics). Users can copy‑paste the code into their own environment to render the graphics.
It would be interesting to have some kind of Lumo‑Python bridge that delivers the generated Python code to a local client. The client runs the code in a sandboxed, offline Python/IPython/Jupyter/... session and returns the rendered image directly to the chat.
This would look like a local client chat application communicating with proton servers, able to execute Lumo-generated Python code on the user’s machine. No data leaves the device, preserving confidentiality and complying with privacy‑first principles, and a sandboxed Python would prevent from possible side-effect. Users would then be able to get visual answers from Lumo in both an efficient and private way.
This may be implemented as an independent client app, or (maybe even better) as an OS-agnostic jupyter kernel for taking advantage of the notebook interface (with something that would be essentially a notebook-integrated Python API to Lumo).
When users request visualizations, Lumo frequently replies with Python code snippets (e.g., matplotlib graphics). Users can copy‑paste the code into their own environment to render the graphics.
It would be interesting to have some kind of Lumo‑Python bridge that delivers the generated Python code to a local client. The client runs the code in a sandboxed, offline Python/IPython/Jupyter/... session and returns the rendered image directly to the chat.
This would look like a local client chat application communicating with proton servers, able to execute Lumo-generated Python code on the user’s machine. No data leaves the device, preserving confidentiality and complying…
2 votes -
one click scroll down to most recent messages (mobile and desktop)
Currently you cannot easily scroll to the most recent answer in a conversation, it's particularly difficult on mobile. The app opens a conversation at the top, on the first query posted to Lumo. You can click the down arrow icon which scrolls partially but not to the bottom necessarily. So for very long conversations it often requires a lot of clicking or dragging of the scroll bar to get to the most recent message in conversation.
Even if you enter a new query it doesn't automatically view the most recent answer (there's a separate User Voice idea for that). Instead it would be ideal if there was a way to easily access the most recent messages in conversation without having to manually scroll. Just click and it takes you to the most recent answer somehow.
Currently you cannot easily scroll to the most recent answer in a conversation, it's particularly difficult on mobile. The app opens a conversation at the top, on the first query posted to Lumo. You can click the down arrow icon which scrolls partially but not to the bottom necessarily. So for very long conversations it often requires a lot of clicking or dragging of the scroll bar to get to the most recent message in conversation.
Even if you enter a new query it doesn't automatically view the most recent answer (there's a separate User Voice idea for that). Instead…
1 vote -
Unlimited Calendar Sync Size
I was hoping to switch to Proton calendar as my "source of truth" across the 2 Google and 2 Office 365 calendars I have to juggle on a daily basis. I am however unable to use it because of the "Calendar Too big" sync limitation.
Even the hack-fix of manually adding "start-min=2025-11-01T00:00:00Z" to the target calendar URL does not work. In fact the form field to paste the URL in is too short in number of characters to accept the full string like this:
This is silly and a little ( a lot actually!) embarrassing.
I am sure Proton has the technical chops to do the software right.
I am sure that Proton does want to be a fit for "Power / Professional Users"
I have no problem paying extra to make this available to me.Please fix this - it is a bad look...
I was hoping to switch to Proton calendar as my "source of truth" across the 2 Google and 2 Office 365 calendars I have to juggle on a daily basis. I am however unable to use it because of the "Calendar Too big" sync limitation.
Even the hack-fix of manually adding "start-min=2025-11-01T00:00:00Z" to the target calendar URL does not work. In fact the form field to paste the URL in is too short in number of characters to accept the full string like this:
This is silly and a little ( a lot actually!) embarrassing.
I am sure Proton…
1 vote -
Lumo prompt-library
Create and save prompt to reuse them with different parameters.
1 vote -
Calendar, sélectionner / désélectionner tous ses calendriers d'un coup
Bonjour,
Dans le calendar, ça serait pas mal de pouvoir sélectionner / désélectionner tous ses calendriers d'un coup
d'avoir une liste de ses évènements et que le "aujourd'hui" soit plus visible sur un affichage mois ou semaine.
merci2 votes -
Lumo crashing on Android with No Google Play Store
Hi team. great work on launching lumo. on my android phone with no google playstore and only google play services running with limited access and internet access blocked by firewall, the lumo app crashes after a delay of a few seconds.
The error reads
'Something went wrong.
Check that Google Play is enabled on your device and that you're using an up-to-date version before opening the app...'all other proton apps work fine on my phone
2 votes -
Lumo Branding / Design
Lumo Logo / Design / Branding
Suggesting an overhaul and alignment with the Proton family of product.
The current choice has no cohesion with the established design language.
Unless it is design for an audience of kids and adolescents specifically it feels
playful, cute and disjointed.Surely kids and adolescents are not the primary market for the company's
flagship AI product. Given the importance and relevance of AI a bit more
thought could very well go into the branding here.As huge Proton fan and customer of many years I frequently wonder about
the design choices. Form, function and aesthetic are frequently not at the level
of the famed Swiss design standards one would expect.Lumo Logo / Design / Branding
Suggesting an overhaul and alignment with the Proton family of product.
The current choice has no cohesion with the established design language.
Unless it is design for an audience of kids and adolescents specifically it feels
playful, cute and disjointed.Surely kids and adolescents are not the primary market for the company's
flagship AI product. Given the importance and relevance of AI a bit more
thought could very well go into the branding here.As huge Proton fan and customer of many years I frequently wonder about
the design choices. Form, function and aesthetic…1 vote -
I would like the option to voluntarily share longer chat transcripts for model training when they contain useful corrections or insights. Wh
I would like the option to voluntarily share longer chat transcripts for model training when they contain useful corrections or insights. While the existing thumbs‑down feedback works for isolated messages, many improvements emerge only after a series of exchanges. Please add a simple opt‑in mechanism (e.g., a checkbox or toggle) that lets users grant permission for these extended conversations to be used for training, while keeping the default setting to keep all other data private.
1 vote -
Add “Auto‑Enable Web Search” Setting for Lumo Chats
Introduce a new option in Lumo’s Settings—“Auto‑Enable Web Search” (or a similarly named toggle). When activated, Lumo would automatically turn on the web‑search capability at the start of every new conversation, eliminating the need for users to manually enable it for each chat session.
Why This Matters:
Seamless Experience: Users who rely on up‑to‑date information can get instant, real‑time answers without an extra step, making interactions feel more natural and efficient.
Consistency: Guarantees that all chats benefit from the latest data, especially for topics that change frequently (news, weather, stock prices, etc.).
Time Savings: Reduces repetitive actions for power users and those who frequently switch devices or start new threads.
Clear Control: The toggle remains user‑controlled, allowing privacy‑focused individuals to keep web search disabled if they prefer a purely offline Lumo experience.
Suggested Implementation DetailsDefault State: Off by default to preserve the current behavior and respect user privacy expectations.
Indicator: Show a subtle icon or badge in the chat window when auto‑search is active, so users know the feature is enabled.
Per‑Session Override: Allow users to temporarily disable web search for a specific chat via the existing manual control, without affecting the global auto‑enable setting.
Potential Benefits
Improves user satisfaction for those who regularly need current information.
Aligns Lumo’s functionality with the expectations set by other AI assistants that automatically leverage live data.
Encourages broader adoption of Lumo Plus, where web search is a premium feature, by showcasing its convenience.Introduce a new option in Lumo’s Settings—“Auto‑Enable Web Search” (or a similarly named toggle). When activated, Lumo would automatically turn on the web‑search capability at the start of every new conversation, eliminating the need for users to manually enable it for each chat session.
Why This Matters:
Seamless Experience: Users who rely on up‑to‑date information can get instant, real‑time answers without an extra step, making interactions feel more natural and efficient.
Consistency: Guarantees that all chats benefit from the latest data, especially for topics that change frequently (news, weather, stock prices, etc.).
Time Savings: Reduces repetitive actions for power users…
2 votes -
starting over from the beginning is annoying
It's very annoying to have to scroll to pick up where I left off in a long chat. Please revert to the way things were before or add a setting for those who want to start the chat from the beginning.
1 vote -
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)
1 vote -
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…
1 vote -
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…1 vote -
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.
1 vote -
New Zealand Calendar Public Holidays.
Add New Zealand as an option in calendar to add public holidays.
1 vote -
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…
1 vote -
In Lumo, re-label the entry box as "Question or Response", as it has to be used to respond to choices offered by the search.
In Lumo, re-label the entry box as "Question or Response", as it has to be used to respond to choices offered by the search.
2 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.
1 vote -
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.
1 vote -
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.
1 vote
- Don't see your idea?