🚀 Introducing beemine.ai - AI-powered Beeminder companion

[troylar]

Hello Beeminder community!

I’m thrilled to announce the launch of beemine.ai, an AI-powered companion designed to enhance your Beeminder experience.

About beemine.ai:

beemine.ai is an AI-powered platform that integrates with Beeminder to provide personalized insights, goal optimization, and enhanced motivation for achieving your objectives.

Key Features:

• Goal Analysis: AI-powered analysis of your goals and progress patterns

• Smart Insights: Personalized recommendations for goal optimization

• Progress Tracking: Enhanced tracking with AI-generated insights

• Beeminder Integration: Seamless connection with your existing Beeminder account

Current Status:

• Beta version fully functional

• Extended free access

• Active development and community feedback integration

• Regular updates and improvements

• Community-first approach - focus on user adoption and value demonstration

Why I Created This:

As a dedicated Beeminder user, I wanted to build something that would help others in the community achieve their goals more effectively. The AI component provides insights and recommendations that can help optimize your goal-setting and tracking strategies.

Community Focus:

This is built by the community, for the community. I’m taking a community-first approach, focusing on building a strong user base and demonstrating clear value before any monetization. Your feedback and suggestions are invaluable and will directly influence the development roadmap.

Try It Out:

Visit beemine.ai to sign up for the free beta. I’d love to hear your thoughts, feedback, and suggestions!

What’s Coming:

• Extended free beta period

• Regular updates based on community feedback

• Community-driven development and features

• Focus on user adoption and value demonstration

• Mobile app development

• Additional integrations

Questions & Feedback:

I’m here to answer questions and listen to feedback. This is a community-driven project, and your input is crucial for its success.

Thank you for being an amazing community, and I hope beemine.ai can help you achieve your goals!

Best regards,

Troy

---

This is an independent project built by a Beeminder community member. Not affiliated with Beeminder.com but built with love for the community.

[scarabaea]

Can you share any examples of what it can be helpful with? So far, it seems like I am not getting it, where should I click to get anything useful. In response to one of the pre-entered prompts “summarize my progress” or something to this effect, for example, I got this most generic ramble to the effect of “you are making progress but there’s always room for improvement”. So I am guessing I didn’t make the inquiry specific enough to what this tool can be helpful with, and I would appreciate some guidance.

[scarabaea]

Some more feedback from clicking on the “provide insight” button by individual goals on the dashboard.

1. Many of us just name the goals with slugs, without providing a narrative name. Sometimes small print when the detail is necessary. This tool only looks at “name” field and screams that a goal is not clearly defined if the field is empty.

2. It doesn’t understand that there are whittle down goals. Maybe it should be programmed to look at the type of goal for this context. Here’s, for example, what it “suggests” me about one whittle-down goal of mine:

Every judgment if wrong here.

With a different goal, it seemed to understand that I am dealing with a backlog and the nature of the goal (reading books in my TBR) but it’s recommendations are preposterous:

3. It seems to insist on having a “clear target” - like reaching 500 hours of language study or a specific amount of vitamins consumed (just two examples from the goals where I had a name explicitly assigned, so the tool understood what the goals are about). Well… I hope I don’t need to explain this is simply not how it works.

[scarabaea]

And a couple of minor UI things:

1. When adding filtering condition by tag, there is not way to choose the tag in question.

2. In quick filters, it would be nice to have tag sorted alphabetically.

[dreev]

Troy, thank you so much, Beemine is bee-yootiful!

(Troy and I have been chatting about this extensively. Troy is amazing and this 100% has our blessing and endorsement.)

And huge thanks to @scarabaea for diving in! Her first note about goalnames vs goal descriptions was the first feature request I submitted on the site. I realize that just putting the goalname in the description is a workaround. Also I think this is our fault for misnaming the fields in the API (“slug” and “title” instead of goalname and description).

[scarabaea]

One more unexpected observation:

The “AI insight” clearly has no idea what my units are for this goal only “presumes” that they are per day. (I understand sometimes it will be necessary to hypothesize, but in this case these hypotheses about “minutes” or “sessions” also make no sense). In the meantime, all of these details are clearly stated for this goal; it seems like the AI hasn’t been pointed to where these contexts can be taken from:

I am sorry if my comments are too blunt. I really think a tool like this could be fun; I personally have a lot of scripts that I am running on my beeminder data to get additional insight into my progress, and I believe adding AI to the process can bring some interesting results. Which is why I dived in to test the tool in application to my goals, but currently it seems like the connected AI doesn’t get enough information about specific goals and doesn’t have enough context about what people use beeminder for in general.

[troylar]

THIS IS FANTASTIC FEEDBACK!

I’ve been a developer for 40 years and I don’t have any notions of grandeur.

This is YOUR (Beeminder community), not mine.

I welcome any feedback, no matter how harsh!

I LOVE the feedback on the AI analysis. I’m going to tackle that today . . . to me, that’s the biggest value.

There are SO many directions we can take the AI in terms of analysis.

There are so many different use cases for why and how people use Beeminder, so I didn’t want to create too much bias.

But I just needed a starting point just to get the conversation started and you’ve given me really actionable feedback already!

[troylar]

Beyond just “how well am I doing on my goals?”

How about also “how well am I TRACKING my goals?”

And "How well do my Beeminder goals align with my current goals?"

In other words:

• What goals am I neglecting to track consistently?
• Are there goals that are derailing frequently and why?
• Should any of my goals be deprecated?
• Are goals configured incorrectly for what I actually need to track?

For me, Beeminder is more than just a tracker.

It’s a canary in my discipline coalmine.

I can measure the quality of my discipline by the accuracy of my Beeminder.

I’ve literally sat by and watched goals derail in slow motion like I’m watching a car crash.

How can I not only get ahead of that, but how can I better improve my life/goals to keep that to a minimum?

[philip]

@scarabaea

The “AI insight” clearly has no idea what my units are for this goal only “presumes” that they are per day. (I understand sometimes it will be necessary to hypothesize, but in this case these hypotheses about “minutes” or “sessions” also make no sense).

I’ve found for my own AI tooling that it’s helpful to put more detail in the fine print about what this goal represents, how best to nudge me to make progress on it, etc. Otherwise it makes loads of assumptions that don’t apply and are often exactly backwards.

[troylar]

BeeMine AI Vision v0.1.0

Release date: 2025‑09‑21
Live at: https://beemine.ai
Changelog: beemine.ai - AI-Powered Goal Tracking for Beeminder

TL;DR

• More intelligent, inline, scan-friendly insights for each goal.
• One-click “Expand” for longer reads.
• Copy insight to clipboard in one tap.
• Better labels, filters, and readability across devices.

What’s new

Added

• Dashboard · Insights — AI Insights now show up in a handy popover with clear, readable formatting.
• Dashboard · Insights — A quick info strip shows units, cadence, and type right above the insight.
• Dashboard · Insights — One-click “Expand” for long reads in a larger view.
• Dashboard · Insights — Copy the full insight to your clipboard in one tap.

Changed

• AI Analysis · Insights — Insights speak your goal’s real units and cadence (no guessing!) and give concise, actionable tips.
• Insights · AI Analysis — Improved readability of insight content and spacing.
• Dashboard · Insights — Smoother, more responsive popover and dialog sizes across devices.
• Dashboard — Goal labels are clearer: slug first (bold), title second (subtle), for faster scanning.
• Dashboard · Filters — Quick Filters: Tags are sorted A→Z for easy selection.

Fixed

• Dashboard · Insights — Scrolling inside insight popovers/dialogs is smooth and doesn’t move the page.
• Dashboard · Accessibility — Improved accessibility and labeling for goal actions and details.
• App-wide — General polish and small stability fixes.

Try it in 30 seconds

1. Open https://beemine.ai
2. Head to the Goals Dashboard
3. Open an insight on any goal
4. Click “Expand” for long-form, and “Copy” to grab the text for notes or coaching

Why this helps

• Faster scanning: clear category tags and consistent typography make it easy to find what matters.
• Actionable guidance: insights reference your goal’s real units/cadence and suggest next steps.
• Shareable outputs: copy to clipboard for your journal, coach, or accountability partner.

Thanks for trying the v0.1.0 preview—your feedback will directly shape the next releases!

[scarabaea]

Great! Now that the AI gets the info about the goals’ slugs for context and units and rate units, it offers way fewer weird assumptions about the nature of the goals. It looks like one more thing that it will be helpful to teach is the “good side of the road” (UPD: SIDE! not size). Just so it won’t suggest that I need to purchase more books as soon as possible, for a goal that is supposed to bring me to reduce the number of purchased unread books

Also, what is a “safe bump”? This is something that has probably been taught to the model as part of the context about beeminding, as this term appears in “insights” not for one but for many goals, but I couldn’t figure out what this refers to, in more usual beeminder parlance.

UPD: I found this, it’s in the API documentation, but instead of removing the question that is no longer relevant, I will post here the response, in case anybody else is wondering:

• safebump (number): The absolute y-axis number you need to reach to get one additional day of safety buffer.

[dreev]

Holy cow, @troylar, you are fast. Big improvements already! Thanks for adding the goalnames especially.

Speaking of which, here a couple terminological quibbles, maybe especially for @scarabaea who’s been around forever:

• not “slug”, goalname (we can still say slug if talking about the final bit of the URL specifically)
• not “road”, bright red line
• and I know this one’s just a typo but in case anyone’s confused: good side (not “size”) of the red line, referring to whether you want your datapoints above it, as in a do-more goal, or below it, as in a do-less goal
• “safebump” is an API field that I don’t think we meant to be a user-facing term, but in the API it means the y-axis value (absolute, not delta) that you need in order to get one additional day of safety buffer. I tend to think users should just care about the current daily rate. That’s the amount you’ve chosen to commit to and an amount that will also guarantee you one additional day of safety buffer. If you perfectly skate the red line, your daily rate and the safebump delta will be the same thing. If you did slightly more than the daily rate some previous day, maybe you can get away with slightly less today to bump yourself up by one safe day. But I think it’s better to just think in terms of doing the daily rate amount daily. That’s what keeps you from losing any of your safety buffer. If it’s a beemergency, Beeminder will tell you your exact bare minimum to hit the red line.

PS: Ah, I see you figured out the “safebump” thing as well. I’ll leave my philosophizing about it in case that’s also useful.

[scarabaea]

Right, I have been around forever and I am refusing to unlearn the terms that were around when I first learned beeminder

But if we are talking seriously, in terms of the user-friendliness of @troylar’s tool, it is important that the model isn’t then taught on terms that are used in API and were either never supposed to be user-facing or reflect that now-revised terminology (like “roadall” matrix). (And by taught I mean “provided for context”, unless I am misunderstanding the setup and actual models are trained for the use in this tool.)

[troylar]

I have loved Beeminder for years and one of my biggest challenges has been understanding all the stuff I’m NOT understanding.

I’m guessing the majority of people use it as a habit tracker, but I’ve seen so many different and interesting use cases.

I want Beemine to be coaching tool that cuts through the vocabulary and helps users leverage Beeminder to its fullest . . . to help us be better every day.

My mission statement for Beemine is elementary: “Be(e) better ever day.”

Here’s how I look at it.

The universe works with or without the man-made science of physics.

Physics simply gives us a common language to describe what already exists.

I can know not to drop a glass on the floor without ever needing to know the word “gravity.”

In the same way (to me), Beeminder isn’t about the vocabulary, but rather what the vocabulary is trying to describe.

In that vein, (I’m guessing) new users probably get snarled up in trying to understand the nuances of the terminology (akrasia, red line, ratcheting, auto-ratcheting) . . . I know I have–especially if I haven’t used the UI for a while.

And that’s where I want Beemine to break through the need to understand the nuances of the terms and coach you into building the “right” Beeminder setup for YOUR personal needs and goals–without needing to be a Beeminder expert.

Right now Beemine is a fancy “wrapper” around the current Beeminder interface–honestly just to get a familiar baseline.

Before I start mucking with things, I want to make sure that it’s speaking the right language.

I want Beemine to be a powerful self-analysis tool to help answer questions like:

• How well am I tracking my goals?
• Are my goals setup in alignment with what I want/need?
• Are my goals realistic?
• What goals am I missing?
• What goals should be deprecated?
• What are the trends in my derailing?
• Are there specific days of the week/month/year that I derail on certain goals?
• Are there relational trends between goals that I’m not seeing?

My ultimate goal is to integrate with Google Calendar, Todoist, Intend (Complice)–ALL of my life and busiess tools, and be able to use AI to interact with the entire story of my life.

• “Looking at my calendar, is my focus goal realistic?”
• “I want to start working out on Monday. Let’s setup a 30-minute daily goal and add a Todoist daily reminder to workout at 6:30 a.m.”
• “I haven’t been tracking my food intake goal. Create a couple daily reminders in Todoist to remind me to keep on track.”

I’m really excited about this and I would love any kind of feedback/suggestions!

[redacted]

All the work is done client-side, correct? You give the app your Beeminder keys and your OpenAI keys, they get encrypted in the browser, then the encrypted copy is sent to the server for storage?

I’m checking whether I understood the security section.

[skorytnicki]

fyi there is an endless loop of requests sent to the backend to user_presence and user_sessions.

Honest feedback, why do I have to pass my own AI key?

I developed bui.interestingprojects.net - my focus was charts and dataviz. I would like to see these too in beemine. Also, I shared the way to develop iOS / Android widget somewhere on the forum.

If you’d like, I can share the code, it’s pretty straightforward anyway, data and chats.

The chart or the formatted information you can derive from the usage patterns can improve your prompts. To make the model understand that it’s done in the evenings but the goal name is “morning-routine“ etc.

I wonder if you’ll be able to make money with this project. I think you correctly identify the problem of poor Beeminder UI, but there’s a lot of tools already and we’re all tinkerers, many coming from the IT. It’s a problem, but not a pain. In the past 3 years I’m here, I had more “write” problems than “read” problems (auto trim safety buffer, enter data automatically etc).

My first thought on AI with Beeminder was to give it some autonomy and escape beeminder micro world. Combine sources from my Health app, RescueTime, todo app and then propose tasks or actions autonomously in real time, provide summary of the day etc. The problem here is to get the data (build the integrations).

Good luck!

[troylar]

No, that’s not quite correct. Here’s how API key security actually works in this app:

API Keys are NOT encrypted client-side:

• API keys are sent directly from your browser to the server
• Encryption happens on the server using Supabase database functions (encrypt_api_key)
• The encrypted keys are stored in the database, not sent back to your browser

For AI API Keys:

• You enter them in the settings form
• They’re sent to the server via secure function
• Server encrypts them using user-specific encryption keys
• Only encrypted versions are stored in the database
• All AI requests go through secure edge functions - no direct client-side API calls

For Beeminder Keys:

• Uses OAuth2 flow - you never manually enter keys
• You’re redirected to Beeminder to authorize the app
• Beeminder sends back an access token
• This token is stored encrypted in the database

The key security principle: Your API keys are never stored unencrypted anywhere, and all actual API calls happen server-side through secure edge functions, not directly from your browser. This prevents API keys from being exposed in browser dev tools or network traffic.

So it’s more secure than client-side encryption - it’s server-side encryption with server-side API usage.

Does that help?

[troylar]

fyi there is an endless loop of requests sent to the backend to user_presence and user_sessions.

I’m using this to track user engagement in my own DB vs. using a third-party tracker.

Honest feedback, why do I have to pass my own AI key?

Actually, you don’t.

I need to call this out and make that optional.

Some people are very skittish about that and I want to give that as an option.

If you’d like, I can share the code, it’s pretty straightforward anyway, data and chats.

The chart or the formatted information you can derive from the usage patterns can improve your prompts. To make the model understand that it’s done in the evenings but the goal name is “morning-routine“ etc.

I would LOVE that. My github is troylar (Troy Larson) · GitHub. Thank you!

I wonder if you’ll be able to make money with this project. I think you correctly identify the problem of poor Beeminder UI, but there’s a lot of tools already and we’re all tinkerers, many coming from the IT. It’s a problem, but not a pain. In the past 3 years I’m here, I had more “write” problems than “read” problems (auto trim safety buffer, enter data automatically etc).

My first thought on AI with Beeminder was to give it some autonomy and escape beeminder micro world. Combine sources from my Health app, RescueTime, todo app and then propose tasks or actions autonomously in real time, provide summary of the day etc. The problem here is to get the data (build the integrations).

THAT is exactly my goal!

Beeminder is just one (but a central) piece of the overall puzzle.

I’ve got a new chatbot sidebar coming in a day or two and I’m going to start wiring in integrations.

[scarabaea]

@troylar

In that vein, (I’m guessing) new users probably get snarled up in trying to understand the nuances of the terminology (akrasia, red line, ratcheting, auto-ratcheting) . . . I know I have–especially if I haven’t used the UI for a while.

I absolutely agree with this need. My few attempts to bring some people in my life to beeminder crashed into this problem. But then it is even more important not to create additional jargon (like “safe bump”).

[redacted]

That’s not end-to-end encrypted then. If the server is encrypting the data or making the API calls, then the server (you) can read the data. To make the API call on the backend, you have to be able to decrypt the API keys. The data being encrypted at rest is enough security, sure, but the specific claims on the front page of “end-to-end encryption” and “Even we can’t see your encrypted data,” are false.