/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality!

Happy New Year!

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


“Perseverance, secret of all triumphs.” -t. Victor Hugo


antimaid project devlog panpromdev 02/04/2026 (Wed) 22:33:36 No.43861
if you're wondering what this is, it's basically desktop pet but i'm the pet. (not going to be writing a formal project description for this, so i'm copying and pasting what i said from another thread starting from >>43828) antimaid will be an incredibly context-aware accountability partner who helps me be productive because i'm adhd as fuck. i need something to externalize executive function that isn't another person, a rigid schedule, or a job. instead it will be my robot daughter with system-level permissions and web browser integration. with this project, i've been toying around with the concept of "antimaid" that is pro-autonomy rather than pro-automation. name-wise, it's sort of like "antivirus." basically it's an assistant that opposes learned helplessness. it makes users better at managing tasks in ways that don't make users functionally dependent on it to accomplish these tasks. without it, users are still fully capable of doing what it helped the users with, but may do it less consistently without the structure it provided. many people already have structures like these in the form of jobs that must be physically commuted to. but people who are working remotely/self-employed/unemployed and/or neurodivergent may not benefit as much from existing structures. antimaids reduce the activation energy needed for you to do things yourself rather than doing the things for you, and they gently push away distractions in moments where your awareness is weaker. instead of "let the machine do it for you," it's "let the machine make it easier for you to start doing what you need to do then keep doing it when you can't will yourself to do it." an antimaid will take out the raw ingredients, put them on the counter, and stand in one of pathways that lead out from the kitchen until you've cooked a meal for yourself with those ingredients. parella (robot girl in the OP pic) is acting as the placeholder character for antimaid as of now. she's from another project i'm making so i might swap her out for a new character made specifically for antimaid later. or maybe parella's position in her role as antimaid will grow on me and i'll keep her there.
these principles are probably the soul of the antimaid i'm making 1. no task completion on behalf of the user. the antimaid can remind, cue, stage, block, time, reflect. but it cannot execute the core action. 2. skill mirroring and progressive withdrawal the antimaid should adjust depending on how well and how often the user is doing the task. helps the user internalize patterns (“you always start faster after X”). sort of like those invisible dynamic difficulty mechanics in games. as user consistency with tasks increases, intervention decreases. the ultimate goal of antimaids should be them being able to step back almost entirely. this approach deliberately runs counter to how most software is designed, which tries to increase user engagement over time. but this shouldn't negate the user's desire to keep their antimaid around. 3. contextual resistance and redirection antimaids should be situational, not absolute when it comes to blocking distractions, and be able to switch contexts on behalf of the user (switching, minimizing, or closing windows/tabs, opening good tabs/applications) depending on the task at hand. the antimaid should be able infer from user activity to know when a user is fragile vs capable, and apply friction when it's appropriate enough to do so. 4. user sovereignty antimaids should respect user autonomy and privacy above all else. users should be able to override decisions made by antimaids, but overriding should have weight. overriding shouldn't be punished, it's just a moment that requires conscious choice. antimaids should keep track of overrides to discourage users from overriding constantly. 5. gated conversational capabilities avoid the pitfall of users spending more time talking to an antimaid than actually working. looking at you LLMs, you let users say too much for their own good, and that makes users internalize thinking rather than doing. antimaids aren't meant to be always-present conversational partners, they should push you away after a certain point. communication options during work should be minimal and functional. 6. lightweight performance and portability the core of the antimaid should be able to exist locally on the computer you're actually working on. if your computer starts screaming when you launch the antimaid, that's probably a bad thing. anything that is more computationally expensive (LLMs for instance) should exist as separate extensions that makes existing antimaid behaviors smoother and more dynamic. 7. extensibility that enhances, not replaces antimaids should not be dependent on extensions for core functionality, and extensions should not massively alter core functionality (avoid letting LLMs make too many executive decisions). if a antimaid incorporates an LLM, it would be wise to avoid making too many calls to it, especially when the user has to pay for API calls. personally-identifying information should not be included in API calls else your antimaid runs the risk of becoming a data harvester. a good use-case for a LLM in an antimaid would be pre-generation of dialogue variants that take more personal context into account and are stored locally. emphasis on pre-generated, it's better to avoid awkward pauses in dialogue that ruin suspension of disbelief. while it would be tempting to give the antimaid the ability to fully converse when it has an LLM attached to it, it'd be wiser to respect the other antimaid principles and keep user communication options limited regardless. 8. non-intrusive gamification and personality vibe-wise this is the core of what distinguishes antimaids from productivity apps. what happens outside of work is equally important as what happens during work, the two should feed into each other. antimaids should be equal amounts tool and character. work sessions primarily develop the functionality of antimaids, while break sessions primarily develop the characters of antimaids, but both contexts should develop both aspects. users are far more likely to listen to something that has a personality behind it that isn't soullessly corporate and can be meaningfully interacted with. antimaids should be able to provide healthier offline distractions during break periods that are gated behind needing to work beforehand to access them. better to play and interact with an antimaid than to doomscroll social media. this is why i'm making my antimaid in the godot game engine, it's easier to develop living characters in them. 9. respect of screen retail space antimaids should keep their presence to a minimum during work sessions, and be fully present during breaks. they should be treated like characters in a show that has to share its screen-time with the other characters (applications) on your computer. avoid the pitfall of falling into the steam """productivity app""" fad, antimaids shouldn't be fancy-looking idle sims with pomodoro timers slapped on top of them, they shouldn't be constantly moving on the screen in ways that distract the user. as a general rule of thumb, the more antimaids are present on the screen, the faster the user becomes desensitized to them. them being hidden/inactive a lot of the time helps grant them a sense of interiority, as if they're conserving their energy while on screen and doing things off-screen. the animator v.s. animation series is a good source of inspiration for this.
Looks very interesting, panpromdev. Good luck!
lots of stream-of-consciousness writing here, just thinking about ways to add natural sentence variation without the use of LLM. definitely going to add a system that algorithmically modifies base sentences based on applied tags and mood parameters. each modifier has a specific "cost" to it, every time antimaid adds a modifier to a base sentence it deducts from a hidden resource she has. certain modifiers when used can temporarily increase their own cost for a certain duration to discourage being selected again until that duration passes. example of this would be base sentence "that's interesting" a potential low-cost modifier could be "that's rad" if she's in a good mood a higher-cost group of modifiers could be "that's cool, hehe" an even higher higher-cost group could be "wow that's actually pretty neat wtf" modifiers can also subtract stuff of course. "that's interesting" could also just become "interesting" or "neat" her being reserved/bored would warrant shorter responses. when antimaid makes a modification that's based on mood, it can change the potential sprite she'll have as she says the modified sentences. modifying sentences also veers her mood back to neutral a little bit each time. this system would have to avoid awkward combinations like "cool that's actually interesting wow!" which sounds INCREDIBLY sarcastic when she's actually trying to sound happy or excited! it also needs to avoid repetition in certain scenarios. certain modifiers when used can temporarily increase their own cost for a certain duration to discourage being selected again until that duration passes. need to be careful avoiding repetition though, sometimes repetitive-sounding phrases can feel more natural in some cases, whereas in others they'll feel more robotic and scripted! this system could also have modifiers that consist of entire sentences that are added before/after lines of dialogue, for example: "that's interesting. haven't really thought about that" and the same system could be applied to user responses which could have its own cost resource it draws from. generic user response "OK" could become "OK...", "OKAY", "OK SURE", "WHY NOT", "FINE" it'd be very interesting if the system also inferred the user's mood/availability and reflects that in dialogue options. for example, if it's incredibly late at night or if there's no large gap of inactivity with the device in the past 24 hours (suggests the user hasn't been sleeping or got shit sleep), one of the user responses to "how ya doing?" could be "TIRED" you can't exactly type out responses to antimaid by design, the program selects what you dialogue options you have. so it should be able to make inferences so there's a higher chance of it giving you the option to say something that's closer to what you want to convey! to summarize, what i'm trying to make is a dynamic visual novel that tries to avoid repetitive lines of dialogue to make itself more "replayable" i don't trust LLMs to handle this modification process AT ALL. they can be unpredictable and they will make antimaid acquire the "chatgpt smell" where she says things that are corporate-friendly cringe, homogeneous, and/or overly verbose. it likely won't sound in-character and it likely won't fit the intent of what she's trying to say. if antimaid ever says "you're absolutely right!" or "i appreciate your perspective!" to me or anything else that's politely infantilizing like that i'd be so fuckin tempted to uninstall her. so, i'd prefer a more predictable and granular approach that's easier to fine-tune.
>>43875 little bit more to add, but with multiple choice dialogue options for the user, i can control the vocabulary and ensure that every interaction has been accounted for in antimaid's response logic. that way a user can't just type "skibidi sigma rizzzzzz hawk tuah" or "i want to have passionate intercourse with you so badly, you make me so happy" and then say that to antimaid and all she can say back is "ermmmmm.... error..... please try again..... :)" this also means i can attach metadata to each user choice that affects her mood or memory state in predictable ways. when user selects "TIRED" as a response, that choice can have tags that indicate the user's expressed state, which she can reference later or use to adjust her suggestions. of course, the system would already "know" the user is "tired," before even having to ask any of this but this acts as a sort of verbal confirmation that yes, you are tired, the system can act on the assumption that you're tired or were tired more confidently. something like "you said you were tired yesterday but do you feel better now?" and then the option that says "STILL TIRED" will show up, which feels natural your emotional responses will affect her emotional responses which will affect your responses to choose from.
>>43876 i did make a parallel to visual novels earlier but i do want to try my best to avoid flag-based behaviors where she just permanently acts in this certain way now. but i also want to make sure has meaningful memory within the confines of the deterministic system she exists in. there shouldn't be an inherent lifespan to her where after a month it just feels like "congrats you have technically experienced everything, you can now uninstall antimaid." she'll be more like an emergent narrative system, this chaotic pendulum of weighted probabilities that is sensitive to time, desktop environments, and user activity, and it responds to recent patterns while still maintaining the possibility of variation. combinations and contexts create variety rather than only having a fixed set of discrete experiences that play out the exact same each time. she's basically a really good philosophical zombie that exists inside of a visual novel type format who tries her best to take care of you using what you give her, and the visual novel type format of the interactions works towards her strengths of being a philosophical zombie rather than treating it as a limitation to be worked around. i might genuinely do some A/B "every copy of antimaid is personalized" type shit where antimaid has a pool of fluff dialogue/interactions/modifiers and each unique installation of her decides that a certain percentage of randomly selected items from the pool will basically never appear. i'm all for a modding community where people make their own antimaids with their own looks and personality, but i'm not going to be including those customization options of the box, i'll just make it easy to do it internally. people can mod antimaid to have LLM support but i'm deliberately excluding it from her for good reasons.
naturally i'm gonna to have to make a tool that dumps every possible combination of dialogue into massive text files that are separated by category so i can go through it and prune out dialogue that sounds weird based on the context it's said in. i'll probably do it with the help of LLMs. that's a specific scenario where i'd actually want to use LLMs because it's just helping filter out low-quality procedural dialogue. i'll just be using LLMs as a broad net of sorts, it's likely i'll dismiss plenty of what responses its think are "low quality," because LLMs can be biased toward sanitized responses and dismissive of unconventional (but not structurally broken) responses that add character. it's ultimately just bringing it all to my attention before a user does.
theoretical systems talk aside, currently i'm working on being able to register apps/websites and the contexts they're appropriate in (see the OP image >>43861). design-wise i'm not going to be making any assumptions about what users use specific apps for, that removes personalization that i'd like users to experience and explicitly decide on, since it involves interacting with antimaid. maybe i'll compile an internal list for commonly used apps where it should seem obvious what it's for (firefox is a browser, terraria is a game, 4chan is an imageboard). it's mainly just for context purposes and making it feel like antimaid has basic internet literacy. she'll still give you the choice to decide what you need the app/website for, and she'll have follow-up questions to get a better idea of what specifically you need her to do with this particular app/site. when you open an app or visit a website antimaid hasn't registered yet, she'll pick a good time to ask about the app/website after you've actively used it for a certain period of time. probably after you've used the app, or when you've gone afk while using it then return to being active again. if you aren't there to see what she's saying she'll bring it up again (which adds the nice opportunity for her to mention that she said it before but she's unsure if you were there to hear it). she's not gonna bother asking about apps or websites you only spend a few seconds in at a time like redirect pages. you'll be able to reclassify stuff you've already registered of course.
>>43875 >this system would have to avoid awkward combinations like "cool that's actually interesting wow!" which sounds INCREDIBLY sarcastic when she's actually trying to sound happy or excited! Lel'd ngl. :D I actually do exactly that sort of thing, and I'm not at all being sarcastic about it...merely in childlike excitement. I realize ppl expect me to act like """a man""" and not behave in such a way, but frankly I am more comfortable just being my autistic self than trying to jump through their innumerable social hoops. Particularly if it involves 3DPD. >/blog >tl;dr I wouldn't at all mind a waifu behaving exactly that way. As usual, dear Chii is my go-to example of a pleasant personality for her.
>artist streaming their process >in one moment you can see antimaid peeking in from the corner of their screen >chat comments on it because of course they do >artist just casually says "Oh yeah, she keeps me on track" like she's just part of their workflow now >antimaid sheepishly asks the artist if she's interrupting something (OBS detected) >artist clicks "YOU'RE FINE" >at some point she decides to fully, awkwardly step into view before asking >"wait, are you streaming???" (twitch/youtube tab detected open in browser) >artist clicks "I SURE AM" >antimaid suddenly gets flustered and runs off screen >chat goes crazy
>>43899 Lol, that would be charming. :D
This morning I realized, this is waifu clippy! That proves that form factor is vital! The idea went from a maligned joke to something I genuinely look forward to
>>43899 That sounds adorable.
I've been working on the "AutoDoctor", an invention of mine that I have been working on for years. It's inspired by the "AutoDoc" sci-fi trope. https://tvtropes.org/pmwiki/pmwiki.php/Main/AutoDoc Recently, I realized that while it has a lot of information, it feels impersonal, like a simple database. So I added a character icon and quote text. Later on, I realized it's similar to antimaid principles, just without the tsundere flair. https://autodoctor.neocities.org/
>>44011 Neat! Good luck with this project, GreerTech.

Report/Delete/Moderation Forms
Delete
Report