Framer AI AgentsDesign and publish professional sites with AI
Promoted
Toone
The AI workspace for your agentic workflows on macOS
300 followers
The AI workspace for your agentic workflows on macOS
300 followers
Toone is the AI workspace for your agentic workflows: a macOS app where specialist agents run your workflows and routines under your control. Describe a recurring job in plain language; Toone turns it into a routine an agent can run again.
This is the 2nd launch from Toone. View more
Toone
Launched this week
Build, run and debug long-running or complex workflows in natural language. Inspect and debug each step with the visibility and control production workflows need. Edit and resume where you stopped.
OpenAI or Anthropic accounts needed
Predictable and governed AI: Craft workflows and routines with the consistency and observability needed for production, with multimodality supported between steps.
Tip: Use ChatGPT desktop for ++ awesomeness
Start building your routines today at trytoone.com
Free
Launch Team / Built With
Toone
Creating consistent workflows alongside the ChatGPT desktop
RunEvr
@matheus_paranhos1 Congrats on the launch !
Workflow automations are what we all need today =)
How’s the accuracy with long and complicated ones? Does the accuracy stay consistent when there are lots of steps?
Toone
@adana Thank you for words, Adana!
And replying to your questions, yes, they stay consistent! The boundaries are well defined so the agents know exactly the steps to take and what to output! Ideally you want to try to start simple and be composing your workflows, instead of creating a big one. But you can also do that, and if it fails, you can simply stop, fix and return from where it was. Very easy, all chatting with the agent 😄
What type of workflows and automation do you most use?
Toone
Hey there, Product Hunt! Matt here, creator of Toone 👋
I rely heavily on AI in my day-to-day work as an entrepreneur and professional, from content creation and outreach to coding. Toone grew out of a problem I kept running into: How to craft workflows more easily, how to make sure they run well and smooth?
Getting an AI automation to do something useful takes work. Then a new project comes in, your needs change, and... you’re back to reworking prompts, rearranging steps, or building the whole thing again from scratch.
I found myself spending more and more time customising the automations that were supposed to save me time. That’s what I started to obsess over: how do I make these workflows reliable without making them a pain to change?
That’s why I started building Toone: to make complex AI workflows easier to create and adapt, with more control over how they run, more predictable execution, and the visibility to understand what’s actually happening before trusting them with production work.
It's been a huge time saver for me, and I hope it is for you as well 👌🙌
Thank you! and have fun on this friday!
PicWish
@matheus_paranhos1 how exactly does gpt desktop app tie into this vs just running it via web?
the timeline with git feature is actually interesting for tracking agent prompt changes!
Toone
@mohsinproduct Hi Mohsin, true! The ChatGpt desktop app can take control of Toone and be used to craft the routines alongside Toone as well. Toone does have a babysitting functionality built in, but you can also add another layer with ChatGpt to operate as you would. So you have ChatGpt babysitting while Toone babysits locally too. Yes, and I tried to make this pretty easy for non technical people by having the agent set this up and users only see what it changes. Thanks for your feedback!
@matheus_paranhos1 - this looks to be a very interesting project. The UI seems simple and clean, I appreciate you are offering free early access, but I am curious has to what @Toone will cost and what model(s) will you be offering: one-off (life-time) payment, subscriptions or pay-as-you go?
Best of luck with the launch.
Toone
@codeandsea I plan to keep Toone available for free and have a routine/automations marketplace as the core business! Thank you for the interest and support, very relevant question!
@matheus_paranhos1 That's awesome, I will definitely be 'kicking the tyres' when I get some free time.
Toone
@codeandsea Yes! 😄 For people who use AI a lot and have 100$-200$ plans are certainly worth it, because as founder you gotta handle SEO, then for that content generation, outreach in linkedin and crafting these routines are very time consuming, but not when using Toone 😄
Outcome
Toone
@dzaitzow Hi Daniel, thanks for coming by and the support! 🙌
I would say that his product is for many types of people, from researchers that would like to have their research done in a particular and structured way, to solo founders who would like to automatize some of their processes like SEO, Outreach. Coding is a great one that I use a lot, because they have lots of steps, each expect certain types of files to exist and the system validates those steps along the way.
In terms of what we got in the market: Think like this, what we have today is a chatbot that will take on your workflow that you spent a lot of time crafting, they are mostly markdowns, the bot can sometimes follow big workflows, sometimes not. But:
1. You don't know where the workflow is happening.
2. You don't know exactly if at each step of the way it's producing what it needs to be producing
3. There's no validation between the steps, meaning the agent can simply output a file with the wrong structure, or missing
4. multi modality is not really a thing
Hope I answered your questions!
Congrats on your launch @matheus_paranhos1 . The website looks great and the concept looks promising.
I really like the idea of repeatable workflows created once and the visualizing it as a graph is just chef's kiss as it makes things easy to understand for non tech people.
A few thoughts I had:
I like the concept of the AI babysitting the workflow but the video also mentioned watching each task as it gets completed. The thing I like about Claude's agentic workflows is that I don't need to watch it work, I can just specify what I need to get done and how I like it to work so it only interrupts me when it is stuck or when I need to review so I'd probably not like to babysit and watch my own workflow in action. Kind of like a CI pipeline, I want to see where things broke as my tasks might take time to finish. Perhaps you meant it only the first time in the video so ignore if that is the case.
I kind of do some similar things using Claude's md files and I ask it to visualize the workflow in a mermaid diagram. It doesn't look as cool as your UI but it does get the job done, what would you say are some other benefits beyond the UX? I do see the UX as a big benefit for non tech users.
Do we have control about which model your tool will use once I give it access to my Anthropic key? This task probably doesn't need a sophisticated model for deep reasoning as this could have a cost implication.
From a UX perspective:
One thing I'd like is the ability to drag and drop nodes to edit the workflow than chatting again as I feel AI agents sometimes rework everything while dragging and dropping could take me seconds
Adding a voice feature would be great as well not only from an ease of use perspective but imagine for use cases like accessibility and IOT integrations.
I hope this was helpful. Great idea, amazing execution on the website and I'm looking forward to playing with it when it goes live.
Well done and all the best.
Toone
@simplyshawn Hi Shawn thank you once again for the comprehensive breakdown, amazing.
Answering your questions:
1. Indeed, We have two planes, agents threads and system agent threads, one are the ones you talk to, the other, are the ones made for the routines. The way it's behaving right now is that if you start to chat to an agent in a system thread, then it goes to the regular agent threads, so treat agent treads as the ones you interact with. Usually they don't show up once you dispatch a routine. You may focus a thread once selecting a graph node, then you are kinda of 'following' the subsequent threads. I find this the best behaviour because like you I usually don't like to get interrupted by anything agents are doing, though sometimes I like to see where they are. So that's why the browser, and the graphs 😁
The babysit functionality is an outer/higher level thread, so it's the same agent for the entire routine.
But take with a grain of salt this functionality as well, since the agent are mostly looking at the outputs and intermediary files, and eventually they may access a certain thread.
2. Mainly the execution system, the fact that you may pause a parallel execution of agents, fix, resume it.
Other than that it works mostly the same, I even sometimes ask the agents to craft a mermaid for me before drafting.
3. You don't give access to your Anthropic Key, I might implement it if someone requests it, but currently it's using claude code or codex. And yes, when running a routine using it can run different models at different steps, if you want to, usually you have a global model, and in a chat thread you can change it to a particular. Or have agents to use specific models by default (not sure if deployed). All that is enabled.
5. Good tip, I was thinking about this is true. It should be relatively trivial
6. We do have voice 😄 english, portuguese, spanish and some others. option + a to access it, or you can tell an agent to record and he will record not only your mic but everything on your sound as well, good for meetings, for example
This seems useful in heavy subagent tasks. Whats the comparative token usage lets say for claude code multi agents and using Toone??
Toone
@madhur_mittal1 Indeed, I'd have to check on that but multimodality is possible, so you may define models for each of the steps so you can have heavier models for certain tasks and lighter ones for others