Launching today
Kanwas
An open-source brain for your team
1.2K followers
An open-source brain for your team
1.2K followers
For you, your agent, your coworker and their agent. It holds the team's critical know-how, research, decisions and data. But it's not a dead storage. It's a workspace that makes the context workable for humans as well as agents.
Kanwas
@predrag_ristic1 Really like that you’re not trying to force everything into chat bubbles. The industry somehow decided every AI product needs to look like another messaging app and it gets exhausting fast.
One thing I’m curious about though is onboarding for non-technical teams. Engineers usually tolerate messy flexible systems because they understand the power behind them, but operations/marketing/sales teams often need stronger structure.
Have you noticed users naturally understanding how to organize work inside Kanwas, or do people initially create chaos everywhere before finding a workflow? Feels like this kind of product can become insanely powerful or completely overwhelming depending on first-time experience.
Welder
@josh_bennett1 Thats a great point and something we are focusing on with kanwas a lot. The first onboarding + what to do after that.
We worried about it a lot at the start but seeing user usage it seems most of them get it. I think canvas interface kind of helps here because most non technical people have used tools like figma or other canvases for creative work.
That said we do plan to do a lot of educational content around kanwas very soon. To really help you get 100% out of it.
Kanwas
@predrag_ristic1 @josh_bennett1 EXACTLY! this comment warms my heart. Canvas is something we've put a lot of effort into and takes a lot of our focus, but I really love it for creative work where the work doesn't collapse.
also everyone starts differently. Someone goes chaos first and then turn it into structure, others like to create structure first and keep it clean
Kanwas
@josh_bennett1 Thank you for the nice comment and referencing on chat interfaces and our angle!
For quite a while we were struggling if canvas view is the best approach, but for all of us in the team it clicked from the start, and it was even hard to describe what exactly is making us to feel like that.
Happy to see that many people resonate with it, and find it more natural than chat ones. I guess it simulates the way the brain works, and also the flow when we are at the desk with pen and papers.
Kanwas
@josh_bennett1 This is great observation. It's the fundamental tension between order and chaos. And to be honest, it's one of the big challanges our users are facing. What we recommend to more non-technical user or someone who is just getting started with this is to keep it simple. Use the default structure, add the important context documents and grow it naturally because too much irrelevant context or some outdated documents can be often worse than nothing at all for the reasoning and outputs of ai agents.
as a solo founder, my 'team' is mostly just me and a handful of agents. keeping the context consistent across all of them is a full-time job. kanwas feels like it could save me hours of 're-explaining' the product vision to my dev and marketing agents. awesome @johancutych
Kanwas
@vikramp7470 yeah like having the whole thinking in one space where it's accessible by both humans and agents is like super power. you do better decisions + you go faster
Kanwas
@vikramp7470 I often use Kanwas for solo work too. The canvas interface is so much better than antyhing i have running locally. Also the CLI is a great way to give my AI agents a place to give me visibility into what is going on and share their output.
Kanwas
@vikramp7470 exactly! the best part is that you can still use your coding agents to pull the context and let it work on top of the same context rest of the agents have.
How opinionated is the agent? Does it mostly organize what is already there, or does it push back on assumptions and ask questions too?
Kanwas
@cam_eddy we've two modes, Direct that is behaving similar to coding agents, executing.
but the mode that I love the most is "get my brain going", it's made to ask you a lot of questions so it really gets your brain going and makes the outputs sharp thanks to your taste and judgment
Welder
@cam_eddy We are really trying to do something different. Most AI tools will just produce more more and more output. We want to focus on quality instead. Output less, ask more and really think with the user.
Kanwas
@cam_eddy Interesting question Cam! As @johancutych already said we support 2 modes, but regardless of that you can update main instructions and your style and way of working to be more push back oriented
Kanwas
@cam_eddy The agent is something we have invested a lot into. It's opinionated, for example it asks a lot of targeted questions to get as much "soft" context that isn't yet captured to help you make good decisions. Being epistemologically honest and working well with assumptions is something we see as the biggest problem of todays AI agents and we are working on every day to improve.
Slazzer
I am glad this is not another closed workspace where everything disappears into a proprietary database. Markdown/YAML plus Git history makes the product much easier to trust.
Welder
@mahin_makkhy This kind of interoperability was very important to us. We didn't want to vendor lock users to some proprietary format but rather let them work on markdown files.
Nice side-benefit of this is that you can use our CLI tool to export and import files easily
Kanwas
@mahin_makkhy .md files are way to go!!
Kanwas
@mahin_makkhy True! Markdowns are the way to go, and from our learnings agent works the best when it lives on top of it.
Kanwas
@mahin_makkhy I'm glad it resonates with you. We belive the era of heavy close, proprietary tools is comming to an end. Open source and open standards / formats are now important more than ever and we've been building Kanwas with this in mind since the first day.
How do you stop the canvas from becoming another messy place over time? Is there an agent workflow for cleaning up old context and decisions?
Kanwas
@ben_stephenson2 this is always a challenge. The brain is made to self-evolve and clean up. Project are made to be messy. We plan to have more agents soon so you can setup linting agent that goes over workspace and find things that are duplicate, old, and maybe not relevant!
Welder
@ben_stephenson2 yep we are working on a system that is going to help with this. We call it "gardener". It will run every morning and tell you what can should be merged / deleted / is out of date.
Kanwas
@ben_stephenson2 we are trying to find balance as AI is still not in the state of handling the context on its own, and we believe it is very important to keep human in the loop. we are working on "gardener" workflows which should make agent resurface relevant stuff and support this flow in more productive way
Kanwas
@ben_stephenson2 We are thinking about it this problem a lot and we are currently working on something we call a "gardener agent", that automates and simplifies this process.
The idea of treating team knowledge as something “living” instead of static documentation is really interesting.
Feels like the challenge over time is keeping the context actually useful instead of turning into another layer of noise as more humans and agents interact with it.
Kanwas
@munevver_ertuncccc and we really believe that, at least for now, it also needs to be transparent, readable, workable by humans. Thats how you get actually living knowledge base
Welder
@munevver_ertuncccc Yep thats one of the biggest reasons we created kanwas. For ourselves first. Its very easy for context to get out of date. But if you can see it, iterate on it and have a smart agent that gardens it output of your LLMs becomes 10x better
Kanwas
@munevver_ertuncccc I agree, that is definitely one of the things that we want to focus on a lot. AI is still not in the phase of taking care of it on its own, and on the other hand teams don't have time to manage and update it, so we are trying to find a balance of keeping people in the loop by making agent support this flow, until we can bet on the complete solution from ai side