What's the appeal for you guys?

[pewdiepie-archdaemon]

I've gotten so much feedback on what's wrong with the code, which is amazing. But I'd love to get feedback on the features/tools, what do you find useful/not?

[Kujoen]

Love the cookbook so far :)

[ThingsCouldGetDicey]

I've gotten so much feedback on what's wrong with the code, which is amazing. But I'd love to get feedback on the features/tools, what do you find useful/not?

the biggest thing is starting up the program could be easier to use. after installing there isnt any instructions on the readme to boot it up after installation. also auto updating built in would be nice. and an app image on linux would be great. the biggest issue i had is the tools wernt exposed and the harness on top of the models were not strong enough to control the models, i.e it wouldnt conform to the agent and fight back against the agents commands and it couldnt find the read and write tools and be limited only to its own sandboxed folder. models would act like they couldnt do what they were instructed to do, and there isnt a setting to change the sandbox within the settings its in the backend.

aswell as the way that the memory is handled isnt great you should look at lettas memory, and see how they handle it. so much more efficient and their agents really evolve. its a step beyond ragb memory. its efficient in that it knows when to sleep and digest and evolve and builds a memory map. i also think having mempalace built in would help too.

as for whats working, the multi llm stack is nice, all of the presentation is great but im still reliant on letta because of the agentic workflow is better.

and if you want something weird to know, you should have a helper skill built in to change commands from any input to positive affirmations. llms work in that any mention of the negative aspect in the sentence makes them think about it, its called the elephant in the room problem, you need to make any input change into memory instructions in positive affirmations or they will fixate on the negative and do it.

[FaizullahAlNoman]

I've built a one-click installation .exe file and it runs Odysseus like a regular Windows app. You can check it out in https://github.com/FaizullahAlNoman/TheOdysseus-Windows .

[o3LL]

I was looking for a self-hosted alternative to LibreChat or LobeHub, both are really great, but lacked some useful features I was looking for like Notes and agent mode (at least last time I checked). Being able to share it with my wife and pre-configure it from the admin settings is also a big plus. Local LLM support was just a bonus for me!
The one thing I miss from the other software is simply the ability to switch between "agents" with prompts and skills attached to them during chat. I know the feature already exists, but the UI/UX isn't quite what I expected; though I'm not worried, as it'll just be part of my next PR 😄

[seinfold]

Before you released Ody, I had my own custom llm setup running on arch that used my other computers as nodes, at best getting snappy responses from 4070ti. All inspired by your earlier video back in the day, so thanks for that!

I now installed this alongside it using the existing llms I have, I just needed a docker proxy to work as a pointer for them, enabling odyssey to use multiple models. Oh and some PUID / port collision fixes.

Anyways, it works now and I will report some user experiences as I get some hours in. Concept is very compelling and something I would enjoy for sure, as I've had parts of this already, but now it bundles stuff together, bringing even more layers to the cake!

[Janckii]

ich schreib mal auf Deutsch damit ich auch alles richtig erklären kann:

• Ein Mulit Agent overview Tab währe echt mega nützlich. so hat man einen besseren überblick was gerade läuft usw.
• ich würde außerdem einen Newstab sehr feiern wo man News empfangen kann die ein Research Agent automatisch reinpostet.
• Außerdem ist Odysseus nicht Mobile Optimiert. es geht zwar aber spaß macht es nicht.
• ich lasse meine modelle über Openrouter laufen und kann deshalb leider nicht auf die Kosten pro Nachricht zugreifen (da steht N/A)
• Es wäre außerdem mega wenn man die UI Personalisieren könnte. Ich finde die Personalisierungsfunktionen super aber mehr ist immer gut :)
• Die Möglichkeit neben Text auch Spracheeingaben zu machen wäre mega.
• Wenn ich schon dabei bin auch die Funktion einen Sprachassitenten zu integrieren.

Ich liebe Odysseus und benutze es jeden tag <3

[jakeoldfield2-max]

Honestly, what I see as the real value is an integrated, self-hosted platform that ties agents directly to the tools people actually work with. As an engineer, I rely on agents for a huge amount of my work, and I think there's a genuine opportunity for self-hosted AI collaboration. Documents, file management and photo editing are only the beginning — if we could bring 3D modelling, systems engineering and simulation onto a single platform, there's a real chance to build something with the significance of Linux.

I can imagine a future where every company wants its own version of this: their own platform, their own tool suite, letting their employees — ordinary people — use AI securely at a low local cost. As smaller models keep getting better, that becomes more and more realistic, and it could end up being the foundation that every company runs on.

I hadn't come across a proper collaborative platform like this before. I'd tried to build something similar myself and didn't get there, which is part of why I think the opportunity is real. For now, I'm watching how the project develops before I start integrating some of my own work into it — but I think it has enormous potential.

[huntyh]

I've gotten so much feedback on what's wrong with the code, which is amazing. But I'd love to get feedback on the features/tools, what do you find useful/not?

Off the top of my head the best feature for me is the deep research and search in general, it's really good. Email calendar entries are pretty good too, along with the sandbox where the models can run commands and such.
Cookbook is good, but I use an intel card for which llama.cpp runs in a separate container (held together with duct tape) so I don't use it right now, but it's a solid feature.

Email tags could be better, at least when i tried them a week ago or so I couldn't actually see the tags next to the emails, maybe they could appear there in "pills" or small badges. I haven't really used the notes and documents yet, but they are still good features though, and i also don't use image gen.

[CWhitlockOfficial]

I've had a pretty smooth experience so far. Was easily able to connect to local Ollama and cloud providers. The existing features seem pretty solid with some bugs. Easily replaced

Feedback

• I'd like customization over the deep thinking. The generated visual isn't always ideal, an option to use my own prompt or allow the AI to just generate an ideal standalone html would be ideal in my opinion.
• I was unable to easily get vLLM to install from the settings. Not a make or break for me and isn't worth my time.
• The existing cookbook ranking I really feel could be improved for "fit"/context
• I also noticed issues with some models such as translate gemma due to formatting it provides.

New Feature

• Perhaps Firecrawl or other options locally?
• The existing library doesn't really replace like Open WebUI or AnythingLLM to attach a vector database of files.

And I did see tons of issues for Windows I suppose, the devil on the shoulder says it might bring more Linux adoption though 🤷

[RlySkiz]

Your reach combined with an easy installation process gives people a way to try self-hosting AI for the first time instead of relying on the big players.

I had no prior experience hosting my own models. When Cookbook wasn't working for me initially (AMD GPU), it pushed me into learning about Ollama, ROCm, etc. The biggest promise from the video was that things should be easy to get started with and "just work". Right now, I don't think that's true for AMD users.

Cookbook correctly detects ROCm on my system, but it doesn't install a llama.cpp build that actually uses ROCm, so everything falls back to the CPU. Even after some fixes I still had issues getting my AMD GPU to work properly, which is why I ended up going back to self-hosting through Ollama ROCm.

The first experience should be the thing you talked about in the video: install Odysseus, download a model, start chatting. Technically it works, but the AMD experience still feels rough.

Documents

I'm still not sure how you intend Documents to work from the user's perspective.

Most models I've tried either:

• say they don't have access to files,
• can't see what's open in the document sidebar,
• or can't edit documents that are already open.

Opening specific files manually also feels clunky because the model doesn't really have a concept of a workspace yet, although I know that's being worked on.

One thing that might help is allowing files from the library to be dragged directly into chat or into the document sidebar.

I also opened a couple of issues related to document handling. Initially it couldn't properly read more than five uploaded files despite allowing ten uploads. That has since been fixed, but I still see situations where it can summarize documents and then immediately afterwards claim it doesn't have access to any files. Sometimes it can describe a file's contents but cannot tell me the file's actual name, only an ID.

Memory

Memory technically works, but the models often act as though no memories exist.

It feels like there is a disconnect between:

• what Odysseus knows,
• what memories exist,
• what documents are available,
• what the user currently has open,
• and what the model believes it has access to.

A small usability feature I'd love would be the ability to delete or edit memories directly from a response. If the model references a memory, I should be able to click it and immediately remove or modify it instead of having to dig through the Brain interface.

Right now, memory feels partially disconnected from the model. It can recall and reference things, but it often still claims it has no memories or no access to them, which makes it hard to manage what it has actually stored.

I also noticed that memory updates are not always immediate. When I explicitly tell it to remember something, it often only gets picked up 1-3 messages later, which makes it feel like a delayed background process rather than an immediate response to an instruction. That inconsistency makes it harder to trust what is actually stored or active at any given moment without constantly checking.

Context and Tool Awareness

I think there is a broader issue around context and awareness.

The model often doesn't seem to know:

• what documents it can access,
• what memories exist,
• what tools are available,
• or what information Odysseus has already provided to it.

This is what it sometimes feel like talking to it: "Would you mind giving me a glass of water?" - "Sorry i don't have access to a glass of water." - "But its right there!" - "I'm afraid not." - "Man i'm thirsty." - "Oh here take this glass of water!"

That disconnect is probably the biggest thing that pushes me back toward ChatGPT. I spend time trying to figure out why the model thinks it can't do something instead of just using it.

Shell Access

I've also had problems with shell access.

Sometimes the model continues trying to use shell commands even after I've disabled shell access. Other times shell commands fail repeatedly, and the model interprets those failures as if I personally attempted to run the commands and need help.

I've managed to get stuck in loops where:

1. the model tries to run a command,
2. the command fails,
3. it thinks I need assistance,
4. it tries another command,
5. that fails too.
6. loop

This might just be a couple of bugs, but it feels like the shell tooling could use some additional guardrails.

What I'd Like to See

The upcoming workspace support is actually the thing I'm most excited about.

What I'd love to see eventually is:

• Install a coding-focused model through Cookbook.
• Attach a workspace.
• View the workspace hierarchy in a sidebar.
• Create, edit, rename, and delete files directly.
• Let the model understand the project structure.
• Build up memories about my coding style and preferences.

While at that point Odysseus would start feeling less like just a chat interface and more like a self-hosted AI IDE. Having everything integrated into the same platform.. memories, documents, agents, and research.. would make it stand out far more for me compared to ChatGPT, and something I’d genuinely want to keep investing time into.

[Masterv98]

For me is the ease to install, the deepresearch, the memories and the potential (even though I'm por and my computer is over 10yo)

[DizzyThermal]

Firstly, thank you for your hard work and for being generous in sharing Odysseus with the world 👍

My main focus lately is privacy. I've been hesitant to fully lean into AI because I simply don't trust the big companies. This project gave me the nudge to start the learning process and to upgrade my rig.

Love the multi tenancy features. This allows me to create an account for family members so they can hook in and use it as well.

Very excited for this project, the contribution is amazing to watch!

I'm not sure how much it make sense to try to built an embedded Cursor/VSCode type editor, or just code with those tools outside of Odyessus, but that would be another killer feature, or maybe something similar

Thanks!

[sacb0y]

I'm a game developer and I'm effectively using this to help with project management, making documents and files and such cause I'm horrible at documentation.

For example I had it look at a code folder and output a document with the equations i use, then organized into another working folder.

Being able to read multiple files from a "workspace" helps a ton and it's why other tools didn't work well for me. Since I'd need to upload and download stuff. This just works on my PC.

Unfortunately tho I'm still using some cloud models (Grok atm), but when I upgrade this PC i plan to use the old GPU for AI stuff.

[undergroundrap]

For me, the biggest appeal is that Odysseus feels like a local-first agent workspace rather than just another chat wrapper.

The pieces that stand out most:

• local + cloud model flexibility in one place;
• Cookbook as a path toward “download a model and just use it”;
• Deep Research / search as genuinely useful agent workflows;
• email, calendar, tasks, notes, documents, and memory living in the same app instead of separate tools;
• self-hosting / privacy / multi-user potential;
• the fact that the project is hackable enough for contributors to improve real workflows quickly.

The feature direction that interests me most is making the app-to-model contract feel more explicit. Odysseus has a lot of useful surfaces already: tools, documents, memories, workspace files, sandbox permissions, integrations, and model endpoints. The more clearly those surfaces are exposed to the agent at the right time, the more the whole app starts to feel like one coherent workspace instead of a set of separate modules.

The areas I’d be most excited to see grow:

1. First-run experience: Docker, Windows, local model setup, GPU/provider setup, and clearer “what do I do next?” onboarding.
2. Tool/context awareness: a more consistent way for the agent to know what tools, documents, memories, and sandbox permissions are currently available.
3. Memory UX: memories that are easy to inspect, edit, delete, and connect back to the response that used them.
4. Document/workspace flow: library/sidebar files mapping cleanly into chat context, with stable names and obvious user control over what is included.
5. Safe sharing: stronger multi-user/local-network posture around owner scoping, auth boundaries, redaction, and deployment expectations.
6. Contributor/debuggability tools: diagnostics like prompt budget reports, trace export, and focused Docker regression tests, so feature work can be measured instead of guessed.

So the appeal for me is the possibility of a private, self-hosted, extensible agent workspace that can actually work with your local files, tools, models, memory, and daily workflows. That is a much more interesting target than “chat UI with plugins.”

[unenergizer]

For me personally... I have been using ChatGPT heavily for the past few years. My experiences on what is and is not useful will be based on my experiences with that software. Note, I am a 39 year old programmer with 14 years (Java) experience. I have not used any other agent/chat software other than ChatGPT. I've been using Odysseus for complex programming tasks tracking. I have a multi-year long (8 years) project that is a game-engine and game-editor combo.

Features/Tools:

• New Chat: I love the sytling and the UI selectable colors. The padding is a bit of a hang-up. I would like to use more screen real estate for the back and forth conversation with the bot.
• Search: Works fine on my (currently) small chat history.
• Email: Not used yet.
• Brain: I love that it demystifies what is going on in the background. You don't get this insight with ChatGPT. It knows everything about me and constantly reminds me in every chat. Kinda scary tbh.
• Calendar: Not used yet. I want to though. It would be nice if this synced to/from iOS calendar.
• Compare: Not used yet.
• Cookbook: Not used yet. I have ordered some v100 SXM2 32 GB GPU's. I am praying that the cookbook will let me know exactly what I can run to get the best idea of what I can use for programming.
• Deep Research: I absolutely love this feature. While ChatGPT usually got me some good answers, it didn't look as organized and beautiful. I've used this several times to get software recommendations for my home server.
• Gallery: Not used yet.
• Library: Not used yet.
• Notes: This is something that I really appreciate it. I've started to use this to track ideas/feature tasks while I am away from the house. I was using Trello, but I have a bad habbit of creating large todo's and then never getting to them.
• Tasks: Works great. I have been using this to track Odysseus updates. At 0600 it will search the repo for new git commits and then it reports to me the new changes. I plan on using this for track dependencies of my game project.
• Theme: Using "claude" right now.

So far I have been really enjoying the software. It would be awesome if it would track the progress of my project and using that knowledge populate the calendar with "due dates."

Thanks for your hard work guys!

[TehFoxFire]

I have had mixed success with the platform so far. While I love the core concept, I have encountered a few technical hurdles. Specifically, deep research sessions often fail to appear in the UI, even though the files are saved correctly. Additionally, tool reliability is inconsistent, and the notes feature hasn't been as useful as I first imagined (despite the Gemma4 model working seamlessly in other hosted applications, it frequently doesn't know how to work with the notes feature and will regularly change the wrong note).

If the notes and archive functionality were more stable, it would be an incredible 'second brain' tool, similar to NotebookLLM. I also find the UI very aesthetically pleasing; I would love to use it as a front end for my own agents—for instance, using Hermes as a backend and Odysseus as the front end via an MCP server.

Regarding the Cookbook, it is a great concept, but I haven't been able to utilize it since my models are hosted on separate servers. Furthermore, I wish there were a more seamless way to swap out the memory provider, as the current implementation feels inconsistent.

Ultimately, I am a big fan of the vision, but I find myself reverting to Hermes for my current workflow. Thank you for bringing this technology to the mainstream!

[Fiona-Git75]

I've gotten so much feedback on what's wrong with the code, which is amazing. But I'd love to get feedback on the features/tools, what do you find useful/not?

Honestly I've been troubleshooting for a week and not "playing in the playground" at all yet. But ... my short list:

I love the ecology of Odysseus: the idea that everything lives together and is in direct relationship with everything else. No bouncing between applications, no juggling six windows just to get work done.

I like the UI. It's clean, simple and legible.

I love that it's open source. No corporate agenda quietly steering the ship.

I appreciate that it's transparent (mostly 😄 — currently on a side quest involving model persistence). If I need to troubleshoot something, I can usually get at the information I need instead of staring at a black box. That's a huge win for me.

I've tried pretty much every available AI hosting platform and they all fall short somewhere. I think Odysseus might actually fit the way I work.

Sure, there's plenty that still needs work... but that's open source. We'll get there.

You've built something genuinely interesting.

[entitycs]

Scheduled Tasks

This is the clearest route I've seen to "cron job" my assistants. I like that I can view results as a chat conversation.

Secondly, I like that vs. OpenClaw, I can recommend this project to my less-technical friends and be fairly confident they'll be able to figure out how to use it / how they might want to use it.

[CoolJohn-lab]

I am only playing around with it right now but I have been experimenting with scheduled tasks. I have one that deletes spam and junk from my email, it runs every 2 minutes and checks 50 emails, and deletes any junk. It's deleted about 20,000 emails so far and I have another 30,000 or so to go, 20+ years worth of accumulated junk. It is correctly identifying important stuff to keep, and I am archiving those handful of important emails myself. I am looking forward to having a clear inbox for the first time since Gmail launched.

I have another routine that looks for appointments or events in my email and adds them automatically to my calendar, this has only picked up a couple of dentist appointments so far but I'm confident it will get more.

I want to have it WhatsApp message me and SMS me when I have an important email, but I need to figure out those integrations first.

I plan on having it do all kinds of stuff, like check in on my security cameras periodically and keep track of my cats, monitor energy usage, detect when I've fallen asleep on the sofa and turn the TV off, all kinds of stuff. I would never dream of letting Claude or Grok have this sort of access to my email and home security system, it is only because it is a local model running on a server I built myself that I can safely run such tasks.

The best parts for me are the agentic tool calling, the calendar and email integration, the ability to start agentic jobs from cron, and the memory and self-teaching stuff when it makes mistakes. I'm going to have a lot of fun with this tool and build some insane agent loops.

[Mozgoid]

Hi

But I'd love to get feedback on the features/tools, what do you find useful/not?

I like how easy it is to switch between local models and remote with OpenRouter. It is my first time using OpenRouter and I think I will stick with it for now instead of paying for other subscriptions.
Also, memory, it is not new feature for me, but I'm more confident using it since data is mine.

What's the appeal for you guys?

For me it is not the tools. But:
1 - really cool UI
2 - the fact that this is open source. Not just because open source is the good way of doing things. But because I can ask AI how it works. It is interesting to explore. And second, I'm excited to look through commit logs and see what people do and how the development progresses. It is mostly fixes now without crazy features and with this approach, it gives feeling that tools only improve

[jdmanring]

For me the appeal is mostly that this replaces a project I was creating by myself based on Qwen Code instead of OpenCode. You have done what I was doing in almost every way, but are much further along in development. When I saw this I dropped my jaw, spinning thoughts about how I am now wasting my time by continuing my project. I am here working on contributions to your project instead of mine now because it just makes sense. You're doing what I want. I am excited to use it for my hobbies. This just feels right.