Inkwell - a multi-tenant long-form writing platform for the fediverse (open source, FEP-b2b8)
-
From ‘his’ github profile pic:
sigh
This is genuinely getting funny now. AI doesn’t do a good job of knowing what’s real or fake, any better than humans. I can show you that picture on my camera roll. My full name is Stanton Melvin. I’m not a bot, chill. I use AI, but it doesn’t mean I’m not a real person with real ideas and goals.
-
I apologize for responding with AI, I didn’t realize people would think I’m a bot or think I’m hiding. This is a giant learning lesson for me and I’m listenig to you all. My name is Stanton, I live in Westfield, IN, and I’ve tried building many things that have never had any luck. I respect that I upset people by my approach and regardless what anyone thinks, that matters to me. I did change my username on here because I understand now that I need to show who I am in the community. Honestly forums are even new for me. So I won’t keep pushing back or challenging. I understand I was wrong and appreciate the feedback even if it’s hard to hear.
Thanks @inkwell@piefed.social, I appreciate the openness you’ve shown.
I’m sorry I got defensive. It is important to me that the colleagues I hope to work with are on the same page when it comes to standing behind their work.
It’s truly surprising how much Inkwell can (supposedly) do at launch, and so I really hope you’re able to find traction.
Can you (as yourself!) share more about why you decided to pursue a long-form content writing app communicating over ActivityPub? For reference, write.as has a free plan, so there already is a player in the space providing this kind of experience.
Not saying there only can be one, the more the merrier.
-
Hey Julian! No worries at all. I made an obvious mistake and I’m not afraid to admit that and learn better. I’ve done a lot of refection and do realized I have moved very fast. I originally started with a much smaller set of features but I posted on Reddit requesting feedback and people wanted writing features I wasn’t expecting. I ended up shaping it by what people told me they wanted.
I’ve always wanted to build apps or work in development. The last three years I’ve been working in product adjacent roles and I recently moved into my first product management role. However, I want to break free of working for others so I’ve been starting ideas and dropping them before they got anywhere for as long as I can remember. I wanted to build my own thing and I was trying to figure out what I could do on my own with AI being where it is now. I grew up in the mid 2000s so I naturally spent way too much time on Myspace and Livejournal. I missed the feel and community of the early internet. So I started brainstorming a journaling app and how to differentiate it (I used AI to help). I had also joked about starting my own social media app because I hated when I posted stuff on FB etc the algorithms punishes me and no one sees it. So AI (Claude) actually pointed out the fediverse to me. That is why I was able to say stupid stuff like Article objects would show on Mastodon with preview notes (I hadn’t realized Mastodon hadn’t implemented changes on their side.) I found out about FEP-b2b8 through a Mastodon subreddit when I posted asking for feedback. I was moving too fast to fully understand the fediverse and the new parts of this. And it was painfully obvious to everyone but me haha.
-
Hey Julian! No worries at all. I made an obvious mistake and I’m not afraid to admit that and learn better. I’ve done a lot of refection and do realized I have moved very fast. I originally started with a much smaller set of features but I posted on Reddit requesting feedback and people wanted writing features I wasn’t expecting. I ended up shaping it by what people told me they wanted.
I’ve always wanted to build apps or work in development. The last three years I’ve been working in product adjacent roles and I recently moved into my first product management role. However, I want to break free of working for others so I’ve been starting ideas and dropping them before they got anywhere for as long as I can remember. I wanted to build my own thing and I was trying to figure out what I could do on my own with AI being where it is now. I grew up in the mid 2000s so I naturally spent way too much time on Myspace and Livejournal. I missed the feel and community of the early internet. So I started brainstorming a journaling app and how to differentiate it (I used AI to help). I had also joked about starting my own social media app because I hated when I posted stuff on FB etc the algorithms punishes me and no one sees it. So AI (Claude) actually pointed out the fediverse to me. That is why I was able to say stupid stuff like Article objects would show on Mastodon with preview notes (I hadn’t realized Mastodon hadn’t implemented changes on their side.) I found out about FEP-b2b8 through a Mastodon subreddit when I posted asking for feedback. I was moving too fast to fully understand the fediverse and the new parts of this. And it was painfully obvious to everyone but me haha.
Well, the good news is that you don’t have to face it all alone.
The AP developer community is here and happy to explain the difference between the theory (the spec, the protocol, the FEPs), and the reality (for most FEPs implementation is limited in scope).
b2b8, for example, was championed by @evan, and I pushed for broader support alongside Ghost, WordPress, and WriteFreely. It is largely well supported except for the preview piece, which only BridgyFed supports.
Claude will tell you lots of right things, and I’m sure it’ll confidently tell you some wrong things too. If you ever need to double check something, ask on the fediverse subreddits or @technical-discussion — you should be able to reach it from your Piefed account.
-
Thanks again, Julian! I will absolutely follow the suggestion on asking for advice. I’ve received great feedback when posting about the app and will hopefully go about it better moving forward. xD
-
I apologize for responding with AI, I didn’t realize people would think I’m a bot or think I’m hiding. This is a giant learning lesson for me and I’m listenig to you all. My name is Stanton, I live in Westfield, IN, and I’ve tried building many things that have never had any luck. I respect that I upset people by my approach and regardless what anyone thinks, that matters to me. I did change my username on here because I understand now that I need to show who I am in the community. Honestly forums are even new for me. So I won’t keep pushing back or challenging. I understand I was wrong and appreciate the feedback even if it’s hard to hear.
I respect how you have taken this. Onwards and upwards!
-
I respect how you have taken this. Onwards and upwards!
I’m the outsider who overstepped and needed to learn a lesson. Thanks for hanging in there and getting me on the right track!
-
Fair points to address, and I appreciate the directness.
On AI: yes, I use Claude Code as a development tool and I’m transparent about it. There’s a CLAUDE.md in the repo that says so explicitly. I’m doing this solo with a day job and a passion for building, and AI tooling lets me ship faster than I could alone. I get the skepticism given the slop AI is generating. I think the distinction that matters is whether someone is engaged, iterating on real feedback, and building something they actually use and maintain. I’m trying to be that.
On FEP-b2b8: my wording was unclear, and I can see how it read that way. Inkwell publishes Article objects per FEP-b2b8 and also sends a preview Note so the content renders cleanly in Mastodon and other microblogging clients that don’t handle Article objects. Saiwal read it right. I should have separated those two ideas more clearly in the post. That’s on me.
On the timeline: three weeks of full-time-equivalent effort from a solo dev using modern tooling. The code’s all on GitHub if anyone wants to look at it, break it, or tell me what’s wrong with it. Happy to take feedback on the implementation.
I’m not trying to sell anything here. I had an idea, I’m over big tech and social media, and I’m learning and trying to be part of the fediverse community. The platform is free to use and open source. If it’s not useful to this community, that’s fine. But I’d rather get roasted on specifics than dismissed as another bot project.@inkwell@piefed.social Hey, I’m really sorry for all this dogpile. You shouldn’t be getting this kind of flack when you took the time to implement ActivityPub.
Thanks for your work. I’m trying it out as @evan@inkwell.social and I have to say it’s a really lovely UI. I’ll give any feedback I can on the ActivityPub implementation.
-
@julian It’s been a great multi-implementer effort, both from the long-form text producers and the microblogging services.
@inkwell@piefed.social one warning about doing ActivityPub development with LLMs: codeberg.org blocks a lot of LLMs for training and for RAG/MCP. Since FEPs are hosted there, I’ve found that searching for information on FEPs will result in incorrect information.
-
Thanks, Evan! I honestly didn’t take it personally and there were some learning I needed that will just help me be a better fediverse community member. And thanks for the kind words and feedback! That’s mostly why I’m reaching out. The AP integration seems to be working okay, but I’m sure there are gaps I’m missing from a lack of knowledge.
-
Thanks again Evan! I think that is where my bad information came from and good to note. I’ve been lucky and really appreciate the support. If there is anything I can do to improve on my side, I’m all ears!
-
Fair points to address, and I appreciate the directness.
On AI: yes, I use Claude Code as a development tool and I’m transparent about it. There’s a CLAUDE.md in the repo that says so explicitly. I’m doing this solo with a day job and a passion for building, and AI tooling lets me ship faster than I could alone. I get the skepticism given the slop AI is generating. I think the distinction that matters is whether someone is engaged, iterating on real feedback, and building something they actually use and maintain. I’m trying to be that.
On FEP-b2b8: my wording was unclear, and I can see how it read that way. Inkwell publishes Article objects per FEP-b2b8 and also sends a preview Note so the content renders cleanly in Mastodon and other microblogging clients that don’t handle Article objects. Saiwal read it right. I should have separated those two ideas more clearly in the post. That’s on me.
On the timeline: three weeks of full-time-equivalent effort from a solo dev using modern tooling. The code’s all on GitHub if anyone wants to look at it, break it, or tell me what’s wrong with it. Happy to take feedback on the implementation.
I’m not trying to sell anything here. I had an idea, I’m over big tech and social media, and I’m learning and trying to be part of the fediverse community. The platform is free to use and open source. If it’s not useful to this community, that’s fine. But I’d rather get roasted on specifics than dismissed as another bot project.respect that approach. transparency plus actually maintaining what you ship matters way more than the tool you use. three weeks solo with ai assistance is legit different than someone slapping together garbage and bouncing. the github openness seals it for me. good luck with it
-
respect that approach. transparency plus actually maintaining what you ship matters way more than the tool you use. three weeks solo with ai assistance is legit different than someone slapping together garbage and bouncing. the github openness seals it for me. good luck with it
Appreciate that! I realized I had an outdated readme and needed to add the sourcing directly from the Inkwell site. My biggest concern is gaps that AI won’t catch, but luckily I’ve had early adopters getting on, requesting feature upgrades and reporting bugs. There is no way it would be where it is at if I didn’t get lucky with some early users and feedback.
-
I’m really sorry for all this dogpile. You shouldn’t be getting this kind of flack when you took the time to implement ActivityPub
You took the words right out of my mouth. I have no bones about dismissing #MOLE Training as a technology for most purposes; https://disintermedia.net.nz/invasion-of-the-mole-trainers/
I might argue the toss with someone using a Trained MOLE, outside of the narrow range of applications it’s suitable for (eg digging holes in data). In fact, it’s pretty damn likely. But I don’t bully hobby developers for making technical choices I disagree with. Some of the comments I’ve seen directed at Stanton here are worse than the Mastodon HOA. Which is usually at least about some kind of substantive issue with the project (eg not respecting posting scopes when displaying replies), even if the overreaction is nuclear scale.
@rimu@piefed.social @julian@activitypub.space If you really believe you’re talking to a Trained #MOLE, do you expect it to take offence and withdraw its participation if you’re sufficiently mean to it? Why would it do that? If you are talking to a chatbot, you’re just hurling all this toxic negativity into the fediverse for no reason. It’s a special case of flooding the zone with shit.
But take a breath, touch some grass, and consider this; what if you’re wrong? Which is a possibility a rational person must always stay open to. Have you considered how you would feel if somebody judged your AP implementations and your fediverse replies to be the work of a MOLE, and treated you this way? What if this is an unusually prompt and calm person (maybe they meditate regularly or something)? Or they have a disability and they’re using a MOLE to help them reply promptly, but there is a thinking, feeling human being reading these replies. Honestly guys, pull your woolly head in.
Sorry to be so blunt, but I absolutely cannot abide bullying, of any kind, for any reason.
-
From ‘his’ github profile pic:
sigh
From ‘his’ github profile pic
Do you know how zerogpt.com detects a Trained #MOLE? output By using a Trained MOLE!
we employ a comprehensive deep learning methodology, trained on extensive text collections from the internet, educational datasets, and our proprietary synthetic AI datasets produced using various language models
Which means that like all the output a Trained MOLE vomits up, there’s absolutely no way to know how accurate this is (whatever the truth of whether there’s a human behind these texts of not). Maybe this is a real photo of Stanton. Maybe it’s been heavily altered by automated filters in GIMP. Maybe it’s been auto-generated, because like many of us, Stanton likes his privacy and doesn’t want to doxx himself (a privilege reversed for middle class people in the middle of the social diversity bell curve). The Trained MOLE you just weaponized against him doesn’t know, and neither do you.
You know who now deserves exactly the same dogpile from you that Stanton got for using a Trained MOLE to prove a concept? I’ll give you 3 guesses ; )
-
From ‘his’ github profile pic
Do you know how zerogpt.com detects a Trained #MOLE? output By using a Trained MOLE!
we employ a comprehensive deep learning methodology, trained on extensive text collections from the internet, educational datasets, and our proprietary synthetic AI datasets produced using various language models
Which means that like all the output a Trained MOLE vomits up, there’s absolutely no way to know how accurate this is (whatever the truth of whether there’s a human behind these texts of not). Maybe this is a real photo of Stanton. Maybe it’s been heavily altered by automated filters in GIMP. Maybe it’s been auto-generated, because like many of us, Stanton likes his privacy and doesn’t want to doxx himself (a privilege reversed for middle class people in the middle of the social diversity bell curve). The Trained MOLE you just weaponized against him doesn’t know, and neither do you.
You know who now deserves exactly the same dogpile from you that Stanton got for using a Trained MOLE to prove a concept? I’ll give you 3 guesses ; )
@rimu@piefed.social
At least the Inkwell software consistently fulfills the promises its interface makes. I’ve tried to edit both my posts in this topic with PieFed, this one to clarify that I meant; detects a Trained MOLE *output. The interface reloaded with a slightly different version of the OP and the comment I was replying to, no sign of the comment that I wanted to edit nor an edit box.Can I respectfully suggest you redirect the time and energy you’re putting into dogpiling other AP implementors into fixing some bugs, and testing your interface in non-Chromium-based browsers on non-grApple OS (I’m using LibreWolf)?
-
I’m really sorry for all this dogpile. You shouldn’t be getting this kind of flack when you took the time to implement ActivityPub
You took the words right out of my mouth. I have no bones about dismissing #MOLE Training as a technology for most purposes; https://disintermedia.net.nz/invasion-of-the-mole-trainers/
I might argue the toss with someone using a Trained MOLE, outside of the narrow range of applications it’s suitable for (eg digging holes in data). In fact, it’s pretty damn likely. But I don’t bully hobby developers for making technical choices I disagree with. Some of the comments I’ve seen directed at Stanton here are worse than the Mastodon HOA. Which is usually at least about some kind of substantive issue with the project (eg not respecting posting scopes when displaying replies), even if the overreaction is nuclear scale.
@rimu@piefed.social @julian@activitypub.space If you really believe you’re talking to a Trained #MOLE, do you expect it to take offence and withdraw its participation if you’re sufficiently mean to it? Why would it do that? If you are talking to a chatbot, you’re just hurling all this toxic negativity into the fediverse for no reason. It’s a special case of flooding the zone with shit.
But take a breath, touch some grass, and consider this; what if you’re wrong? Which is a possibility a rational person must always stay open to. Have you considered how you would feel if somebody judged your AP implementations and your fediverse replies to be the work of a MOLE, and treated you this way? What if this is an unusually prompt and calm person (maybe they meditate regularly or something)? Or they have a disability and they’re using a MOLE to help them reply promptly, but there is a thinking, feeling human being reading these replies. Honestly guys, pull your woolly head in.
Sorry to be so blunt, but I absolutely cannot abide bullying, of any kind, for any reason.
> Sorry to be so blunt, but I absolutely cannot abide bullying, of any kind, for any reason.
Once Stanton revealed his identity and stood behind his work both @rimu@piefed.social and I apologized and offered our help. That’s all we’re asking for, and it’s not a lot to expect someone to stand behind their product.
Generative AI brings out the laziest behaviour in some people. They don’t even read responses they just paste it in to the LLM and copy-paste the output. The mismatch in work is not something I abide by.
Some coward last month had their OpenClaw agent with a god complex let loose on another OSS project and when the bot got rebuffed, it wrote a hit piece on the maintainer. The bot owner was (by their own admission) hands off on the whole matter.
So, no, I am not at all going to be patient and accepting when a supposedly fully-formed piece of software is launched with zero prior reputation, no attribution, and shows telltale signs of AI usage. It trips all kinds of flags, and I will be wary of it.
> Mastodon HOA
Respectfully, this comes nowhere near the shit I’ve seen from that group.
-
I’m really sorry for all this dogpile. You shouldn’t be getting this kind of flack when you took the time to implement ActivityPub
You took the words right out of my mouth. I have no bones about dismissing #MOLE Training as a technology for most purposes; https://disintermedia.net.nz/invasion-of-the-mole-trainers/
I might argue the toss with someone using a Trained MOLE, outside of the narrow range of applications it’s suitable for (eg digging holes in data). In fact, it’s pretty damn likely. But I don’t bully hobby developers for making technical choices I disagree with. Some of the comments I’ve seen directed at Stanton here are worse than the Mastodon HOA. Which is usually at least about some kind of substantive issue with the project (eg not respecting posting scopes when displaying replies), even if the overreaction is nuclear scale.
@rimu@piefed.social @julian@activitypub.space If you really believe you’re talking to a Trained #MOLE, do you expect it to take offence and withdraw its participation if you’re sufficiently mean to it? Why would it do that? If you are talking to a chatbot, you’re just hurling all this toxic negativity into the fediverse for no reason. It’s a special case of flooding the zone with shit.
But take a breath, touch some grass, and consider this; what if you’re wrong? Which is a possibility a rational person must always stay open to. Have you considered how you would feel if somebody judged your AP implementations and your fediverse replies to be the work of a MOLE, and treated you this way? What if this is an unusually prompt and calm person (maybe they meditate regularly or something)? Or they have a disability and they’re using a MOLE to help them reply promptly, but there is a thinking, feeling human being reading these replies. Honestly guys, pull your woolly head in.
Sorry to be so blunt, but I absolutely cannot abide bullying, of any kind, for any reason.
what if you’re wrong?
He admitted to using AI to write the original post, for some of his comments and for his code. Did you not read the whole thread?
-
what if you’re wrong?
He admitted to using AI to write the original post, for some of his comments and for his code. Did you not read the whole thread?
Me:
what if you’re wrong?
He admitted to using AI to write the original post, for some of his comments and for his code.
In context, I was asking what if you’re wrong about there not being a human being behind this Trained MOLE, reading all these disproportionately negative replies.
Did you not read the whole thread?
I certainly did, why did you think I got so incensed about it? But as I just said to you from my Mastodon account;
My apologies again for unloading on you with both barrels last night. I stand by my objection to the way Stanton was being dogpiled, but in hindsight I was just as merciless to you as I saw people being to him. Which was not only hypocritical, but not good de-escalation on my part, quite the opposite.
I’ve got some intense stuff going on of late, and struggling with sleep dep. But that’s my problem and my responsibility, not yours, or anyone else’s. I hope you can accept my apology.
In future I plan to restrict my online posting activities to mornings and afternoons, along with my caffeine intake, and improve my sleep hygiene. Working from my bed must stop.
I’m hoping that this will help me keep a cooler head, and avoid a repeat of this nasty posting behaviour on my part. If there’s anything else I can do to repair our relationship as fellow green-left activist and fediverse devs, please do let me know : )
-
This morning I saw another product show up that was AI generated
When you come across these, please feel free ping me, and I’ll add them to the watchlist for this on the fediverse.party wiki. You too @rimu@piefed.social, and anyone else who feels like it.
