> Fun fact: a chunk of Odysseus was built *from a phone* -- mobile shells (Termux), the PWA install, and on-device agents. So "works on mobile" isn't an afterthought, it's where a lot of it actually happened.
Because I've seen AI do it myself? It will very readily take a lofty aim from the brainstorming phase, maybe mentioned off-hand or out of context in a planning doc somewhere, and regardless of if it's implemented it will claim it's part of the delivered product.
Not hating because his project is very cool but this is just a python UI on top of other oss - I'm only saying this bc team at Princeton and I have been building wthis but ground up in rust with more utility and running more custom tools, app integration, custom kernels outperforming llamacpp much more here www.conifer.build
Is it required to register to try and use your service. If I want local AI it is precisely to avoid having to register to a third party service. What am I missing here ?
"Polishing" is something you do over time, and requires usage over time together with refinements, you can't spend a couple of hours refining the UI and suddenly it's "polished", at least in my mind.
TL/DR - "cleaner and more polished" implies no vibe coded slop, and the only reason people start these projects in 2026 is because they're addicted to vibe slop coding.
The truth is, there's something about the vibe coder hive mind where they all create the same thing. Everyone is working on the same thing. There's hundreds of projects like this where the creator thinks they've created the most unique powerful thing.
I think it demonstrates how overuse of LLMs for brainstorming/planning and building destroys creativity. That zeitgeist changes from month to month, but when a vibe coded "AI Native" project gets released you'll always see dozens of people in the replies talking about the identical project they've also built or are working on.
I also notice this in my own work, the people who have most eaten the LLM brainworm will come to me asking for us to build the same projects. They all eventually converge to wanting some sort of power user app (like this) that can just do everything because AI is magic to them and there's no reason why it can't do everything.
I hate Peter Thiel but he has that famous quote form years ago where he says not to build the obvious thing because the obvious thing isn't special, so maybe this has always been a phenomenon.
I think the thing is users run into the same problems (barely desirable tool UX), and arrive at very similar solutions (but with their own small twists and preferences). And now that the barrier to make things is essentially gone, they jump in and make things that git their own desires week because nothing out there was an exact fit.
Like I found a really cool project a few weeks ago that I started using and even made a few QoL changes to. Then I thought of the many other changes I wanted, and that the project was written in a language I'm not familiar with (Rust), and I started a port in what I know well (Python) so that I'll be able to actually review the code if needed and generally make better architectural decisions.
Yea good point, the branding clause is a real gotcha, technically "open source" but with that restriction baked in, which makes it source-available rather than properly free under the OSI definition. A lot of people don't realize until they try to fork or rebrand it for a team/customer deployment and hit the wall.
Conifer (launching tomorrow, June 1st) goes the other way on this, properly open source with no branding strings attached, so you can actually fork or rebrand if you need to. Local AI runtime + IDE, native on Mac/Linux/Windows: conifer.build
I tried installing openwebui once. I gave up after the first 12 gigabytes of pip packages. Made my own LLM chat UI in 500 lines of HTML/JS. That's really all I needed.
Surprisingly, there's a lot of items I'm seeing at first glance here at Odysseus that open-webui either doesn't have or takes extra effort to add in and I've used the latter for a really long time already.
Stuff like an agent mode, deep research and document work are things you'd be able to have with open-webui as well but Odysseus seems to have them thought out.
I am a 'fan' of Open Web UI, but the document editing mode is a compelling feature that Open Web UI does not have. I'll probably wait a while before trying Odysseus... let the inevitable security problems work themselves out.
Is the person best known for exploiting security vulnerabilities to get into places you'd rather they didn't, and then getting lost and taking forever to get places the ideal namesake for your AI workspace project?
Out of all the homeric heroes, the main characters are just one tragic flaw after another. Odysseus is pathologically distrustful, Achilles is a diva that throws temper tantrums.
You gotta pick a dude like Sarpedon who just was a swell guy poking holes in greeks with his spear until he gets killed by Patroclus and everyone is sad.
I can think of worse. Inspired by somewhat surrealist names of a cat video, I'm pretty sure naming them "HP Inkjet 3482" would be a worse name for an AI model and lead to copyright issues from Hewlett-Packard.
I'm rereading the Iliad right now and was just reflecting on the fact the first time you see Odysseus at work he's selling two incompatible flavors of bullshit to officers and their men, in service of a third flavor of bullshit that was decided upon in advance at a war council.
The thing that intrigues me on this one is the Opencode roots.
I've currently host both Open WebUI and Opencode on a server setup and there's multiple things that Open WebUI has implemented very nicely over time but it was also prior to the emergence of coding agents / harnesses. As such, its strengths is more on the `/chat/completions` era and RAG and its own functions and ecosystem setup before tool calls were a thing.
If this stuff works for you already, I think that's fine. That's not a bad thing, and if anything, they've made attempts to improve esp. with the recent backend upgrades and projects like `open-terminal`.
However, Opencode has been built for the coding agent setup since its early days and Odysseus gives me more confidence to run sessions much better knowing that it runs on an Opencode core.
Stuff like MCP support and a clearer lifecycle for example are things I've never really seen implemented well on open-webui (they've improved and support this now, but I'm biased and am salty of the era from its early days of MCP support when they insisted you place a bridge (mcpo) in between just to make it do MCP.). In contrast, opencode does this very well since what they built that works on the terminal also applies to the desktop app and the ACP support for IDEs that use it to connect to the opencode harness.
At the minimum, I expect Opencode to work wonderfully against the frontier models, and all models they host on their Zen and Go service, so that adds some confidence as well on complex workflows.
In contrast, I still find model management a pain in the ass on open-webui (has always relied on LiteLLM or a third-party inference gateway) just to wrangle the models. Hell, they still tag the Responses API "experimental" today, a format that's been around since over last year.
Does that transfer to Odysseus? Maybe, maybe not. Seems it's based on it but a quick check on code says they merely adapted it so maybe it's the latter. I haven't tried it at depth and idk how well Pewds has implemented it, but if it improves on that angle and gets maintainers that accepts contributions from the community, I feel like it'll do nicely.
sure i can share it. it started as a text editor that i incorporated an agent into then moved to a standalone agent. now i’ve got the editor, agent, and have begun work on actual from scratch models (these aren’t available yet), but i’ve been messing with fine tuning ollama coder models as something in the meantime. the goal is embedded models accessible to everyone offline. too many tasks require many less parameters than people are using on them and don’t need to be routed through data centers. mainly just built it for myself but figured others might find it neat and/or useful. it’s all C, no packages or third party libraries. also to be clear i’m not against ai, im against it making 100% of the decisions for you. outside of its own models it’s capable of OAuth with claude and gh. it can also api into many others. tooling is always ongoing maintenance and i’ve got the next version of basically all three of them on the back burner waiting to be pushed. so anyway, here’s my little setup, here’s to not getting killed in the comments. contributions also welcome.
https://github.com/mithraeums/hako-code
its dense in features but the amount of security concerns and improvements would be too difficult to maintain with the way its organised currently. let's see.
All these inexperienced vibe coders build identical slop UI, but what's funny is they usually think its super unique and are super proud of it. I hate them all.
A person that is expressing something a lot of real people are feeling. AI is churning out this terrible UI / graphic desing that looks the same, gives aways is just vibe coded slop and is VERY ugly.
reminds me of the milla jovovich vibecoded ai memory project that got a little bit of hype but was mostly junk... guessing we're going to see a lot more "developer talked to a celebrity and came up with an idea for their money"
Pewd's video provides more context [1].
I didn't see it linked from the project; it seems pretty relevant given who he is.
[1] https://www.youtube.com/watch?v=rAzT5lcezPs
I liked this part https://www.youtube.com/watch?v=rAzT5lcezPs&t=524s
where he describes sending an AI generated email response as "the most polite fuck you to someone that they will never know".
The original email asked him sth like "how many parameters does deepseek v3 have" that "he could have easily searched himself".
PewDiePie on Hackernews what a time to be alive!
You laugh you lose
> Fun fact: a chunk of Odysseus was built *from a phone* -- mobile shells (Termux), the PWA install, and on-device agents. So "works on mobile" isn't an afterthought, it's where a lot of it actually happened.
https://github.com/pewdiepie-archdaemon/odysseus/commit/0517...
this was... hallucinated? hence the removal?
Why would that be the first thing you think of?
Because I've seen AI do it myself? It will very readily take a lofty aim from the brainstorming phase, maybe mentioned off-hand or out of context in a planning doc somewhere, and regardless of if it's implemented it will claim it's part of the delivered product.
Not hating because his project is very cool but this is just a python UI on top of other oss - I'm only saying this bc team at Princeton and I have been building wthis but ground up in rust with more utility and running more custom tools, app integration, custom kernels outperforming llamacpp much more here www.conifer.build
That looks good - nice to have an alternative too.
that looks nice. Is it open source? The links to Github are not working
no github, tiny screenshots on websites, is this malware?
Is it required to register to try and use your service. If I want local AI it is precisely to avoid having to register to a third party service. What am I missing here ?
Can anyone recommend something similar but cleaner and more polished?
Librechat?
https://github.com/mudkipdev/chat
We are working on Msty Claw that's local, private, and no telemetry as well: https://msty.ai/claw
When was the first release of this? Parent mentioned "polished" so I'd assume something at least a 6-12 months old, but that looks kind of brand new.
Interesting. I would not think of age when someone requests “cleaner and more polished”. My mind translates that to “nicer UI”
"Polishing" is something you do over time, and requires usage over time together with refinements, you can't spend a couple of hours refining the UI and suddenly it's "polished", at least in my mind.
This wasn't possible 12 months ago, and barely 6 months ago.
Alright, so couldn't possibly be polished by now then, thanks!
No.
TL/DR - "cleaner and more polished" implies no vibe coded slop, and the only reason people start these projects in 2026 is because they're addicted to vibe slop coding.
Maybe I am misunderstanding but I feel like https://www.jan.ai/ is similar and works great for me
Lemonade, in particular if you are running AMD hardware due to extra optimization (Ryzen AI series CPUs with integrated NPU and/or Radeon GPUs):
https://lemonade-server.ai/
Why not just use open webui?
Open webui isn't fully open source, because it doesn't allow you to change or remove Open Webui branding.
I don't know if that was a motivation for creating this, but it is a reason some people might choose not to use it.
The truth is, there's something about the vibe coder hive mind where they all create the same thing. Everyone is working on the same thing. There's hundreds of projects like this where the creator thinks they've created the most unique powerful thing.
I think it demonstrates how overuse of LLMs for brainstorming/planning and building destroys creativity. That zeitgeist changes from month to month, but when a vibe coded "AI Native" project gets released you'll always see dozens of people in the replies talking about the identical project they've also built or are working on.
I also notice this in my own work, the people who have most eaten the LLM brainworm will come to me asking for us to build the same projects. They all eventually converge to wanting some sort of power user app (like this) that can just do everything because AI is magic to them and there's no reason why it can't do everything.
I hate Peter Thiel but he has that famous quote form years ago where he says not to build the obvious thing because the obvious thing isn't special, so maybe this has always been a phenomenon.
It's like the Lisp Curse but for everything now.
Blockchain solves this
I think the thing is users run into the same problems (barely desirable tool UX), and arrive at very similar solutions (but with their own small twists and preferences). And now that the barrier to make things is essentially gone, they jump in and make things that git their own desires week because nothing out there was an exact fit.
Like I found a really cool project a few weeks ago that I started using and even made a few QoL changes to. Then I thought of the many other changes I wanted, and that the project was written in a language I'm not familiar with (Rust), and I started a port in what I know well (Python) so that I'll be able to actually review the code if needed and generally make better architectural decisions.
It's personal computing time...
Plus, if you look at OpenWebUI horror of a commit history, they don't often take external contributions.
Yea good point, the branding clause is a real gotcha, technically "open source" but with that restriction baked in, which makes it source-available rather than properly free under the OSI definition. A lot of people don't realize until they try to fork or rebrand it for a team/customer deployment and hit the wall.
Conifer (launching tomorrow, June 1st) goes the other way on this, properly open source with no branding strings attached, so you can actually fork or rebrand if you need to. Local AI runtime + IDE, native on Mac/Linux/Windows: conifer.build
I tried installing openwebui once. I gave up after the first 12 gigabytes of pip packages. Made my own LLM chat UI in 500 lines of HTML/JS. That's really all I needed.
This project does look fun though.
did you share your UI anywhere? would like to just pull a proven one versus vibe code my own.
Wow, if you you could've just done docker compose up. I guess we'll never know.
Surprisingly, there's a lot of items I'm seeing at first glance here at Odysseus that open-webui either doesn't have or takes extra effort to add in and I've used the latter for a really long time already.
Stuff like an agent mode, deep research and document work are things you'd be able to have with open-webui as well but Odysseus seems to have them thought out.
I am a 'fan' of Open Web UI, but the document editing mode is a compelling feature that Open Web UI does not have. I'll probably wait a while before trying Odysseus... let the inevitable security problems work themselves out.
I really wonder if people who make such projects actually use them on a regular basis. I cannot imagine myself working with these UIs.
Does fit the rest of his desktop setup, so I guess he likes that stuff
https://github.com/pewdiepie-archdaemon/dionysus/blob/dionys...
Onyx CE self hosted looks better
Oh god the prompt injection.
This project is fun though! And the newest YT comments are very entertaining.
> Odysseus
Is the person best known for exploiting security vulnerabilities to get into places you'd rather they didn't, and then getting lost and taking forever to get places the ideal namesake for your AI workspace project?
Out of all the homeric heroes, the main characters are just one tragic flaw after another. Odysseus is pathologically distrustful, Achilles is a diva that throws temper tantrums.
You gotta pick a dude like Sarpedon who just was a swell guy poking holes in greeks with his spear until he gets killed by Patroclus and everyone is sad.
thank you for this, and also for your search engine
keep in mind pewdiepie is probably the worst dev name ontop
I can think of worse. Inspired by somewhat surrealist names of a cat video, I'm pretty sure naming them "HP Inkjet 3482" would be a worse name for an AI model and lead to copyright issues from Hewlett-Packard.
Sounds pretty consistent with my experience.
https://www.reddit.com/r/ClaudeAI/comments/1r186gl/my_agent_...
>Claude steals user's API keys then berates user for poor security practices
>Claude (via the subreddit's automod account) shows up in thread and roasts user again
I'm rereading the Iliad right now and was just reflecting on the fact the first time you see Odysseus at work he's selling two incompatible flavors of bullshit to officers and their men, in service of a third flavor of bullshit that was decided upon in advance at a war council.
PewDiePie is living an enviable life in Japan with his wife and kid doing cool projects.
What would make an existing OpenUI/Open WebUI user switch to this?
The thing that intrigues me on this one is the Opencode roots.
I've currently host both Open WebUI and Opencode on a server setup and there's multiple things that Open WebUI has implemented very nicely over time but it was also prior to the emergence of coding agents / harnesses. As such, its strengths is more on the `/chat/completions` era and RAG and its own functions and ecosystem setup before tool calls were a thing.
If this stuff works for you already, I think that's fine. That's not a bad thing, and if anything, they've made attempts to improve esp. with the recent backend upgrades and projects like `open-terminal`.
However, Opencode has been built for the coding agent setup since its early days and Odysseus gives me more confidence to run sessions much better knowing that it runs on an Opencode core.
Stuff like MCP support and a clearer lifecycle for example are things I've never really seen implemented well on open-webui (they've improved and support this now, but I'm biased and am salty of the era from its early days of MCP support when they insisted you place a bridge (mcpo) in between just to make it do MCP.). In contrast, opencode does this very well since what they built that works on the terminal also applies to the desktop app and the ACP support for IDEs that use it to connect to the opencode harness.
At the minimum, I expect Opencode to work wonderfully against the frontier models, and all models they host on their Zen and Go service, so that adds some confidence as well on complex workflows.
In contrast, I still find model management a pain in the ass on open-webui (has always relied on LiteLLM or a third-party inference gateway) just to wrangle the models. Hell, they still tag the Responses API "experimental" today, a format that's been around since over last year.
Does that transfer to Odysseus? Maybe, maybe not. Seems it's based on it but a quick check on code says they merely adapted it so maybe it's the latter. I haven't tried it at depth and idk how well Pewds has implemented it, but if it improves on that angle and gets maintainers that accepts contributions from the community, I feel like it'll do nicely.
It reminds me of the Nord theme https://github.com/nordtheme/nord
Watching pewd's continually evolving AI-era tech journey has brough me a lot of joy.
How does this compare to Hermes Agent?
Whoa! Sounds interesting, I'll be taking a deep dive.
nice
Is that by that PewDiePie?
Now waiting for some VC-backed AI company to pick up PewDiePie like they did with the OpenClaw guy.
Somehow he probably will make millions from this just cuz he has a personal brand, despite it being a pile of slop.
Oh yeah definitely, if he went in for doing software he would make millions for free.
I use Librechat
PewDiePie build a self hosted AI workspace. It looks interesting.
No it doesn't.
There's already plenty of pimono / open code webui / electron wrappers
It will be interesting to watch him deal with hundreds PRs.
"gpt can you do these PRs for me"
Is it called Odysseus because it's always spitting out verbose bullshit?
Open code....
Trash.
Pimono or go home
so... AnythingLLM? https://anythingllm.com/desktop
but with extra steps
Wake me when you can increase the awfully tiny font size of the nav.
i write an agent in C, create my own models. crickets. pewdiepie gets into programming on a whim creates slop, 3k stars.
Something something meritocracy blabla
you forgot step 1. become the #1 youtuber
AI doesn’t have a model or agent availability problem to be fair, it does have a positive outreach problem and pewdiepie can do extremely well there
just my 2c
okay i watched his video on it, i can definitely abide.
Did you share your C agent and models? Would love to checkout
sure i can share it. it started as a text editor that i incorporated an agent into then moved to a standalone agent. now i’ve got the editor, agent, and have begun work on actual from scratch models (these aren’t available yet), but i’ve been messing with fine tuning ollama coder models as something in the meantime. the goal is embedded models accessible to everyone offline. too many tasks require many less parameters than people are using on them and don’t need to be routed through data centers. mainly just built it for myself but figured others might find it neat and/or useful. it’s all C, no packages or third party libraries. also to be clear i’m not against ai, im against it making 100% of the decisions for you. outside of its own models it’s capable of OAuth with claude and gh. it can also api into many others. tooling is always ongoing maintenance and i’ve got the next version of basically all three of them on the back burner waiting to be pushed. so anyway, here’s my little setup, here’s to not getting killed in the comments. contributions also welcome. https://github.com/mithraeums/hako-code
rich of you to talk about slop when your own 'handwritten' project has one c file with over 6.5k lines of code.
did you not read the entire post? also correct the one file IS the entire thing. compiling one file on iSh on an iPhone actually works well.
Are you one of the top 100 most famous people on planet earth?
What the hell were you expecting. Your slop will fade in history... his slop will linger for months, maybe a year or two.
Then no one will care or remember about any of this.
accurate, but i’ll keep building anyway
respect for what pewdiepie's trying to do (as he explains in his video) but this is just ai slop
its dense in features but the amount of security concerns and improvements would be too difficult to maintain with the way its organised currently. let's see.
do you think Codex and Claude Desktop are written by hand? "ai slop" means nothing anymore
God the UI design is atrocious.
All these inexperienced vibe coders build identical slop UI, but what's funny is they usually think its super unique and are super proud of it. I hate them all.
Who are you again?
Who are you again?
A person that is expressing something a lot of real people are feeling. AI is churning out this terrible UI / graphic desing that looks the same, gives aways is just vibe coded slop and is VERY ugly.
Who are YOU?
reminds me of the milla jovovich vibecoded ai memory project that got a little bit of hype but was mostly junk... guessing we're going to see a lot more "developer talked to a celebrity and came up with an idea for their money"
except pewdiepies actually the developer in this equation too, its not just attaching his name to it, hes building the projects
it's pretty clear that AI is the developer in this equation