Awesome! I’ll take a look, thanks
10 karma · joined October 3, 2025
Ultimately, for better or worse, AI will be making purchase decisions and my interest is more about how to sandbox them & make them safe (even though this is quite a rudimentary first pass!)
> Homepage is nothing but "Sign in to continue"
The main homepage leaves a lot to be desired but point noted - I'll update the link to direct there instead.
> Perhaps run an anonymous twin of the entire service too. See how they diverge over time
not 100% sure what you mean here. Could you explain further?
> Expect massive legal and KYC stuff too
Noted. The banking provider forces KYC which users go through on sign up before the account details & Privy managed ( https://www.privy.io/ ) wallet that stores the USDC is all provisioned.
> Maybe don't use a bank account anywhere and just accept direct crypto deposits
This would avoid a lot of legal + KYC but I'm aiming to target "power users" who don't necessarily need to know how to on-ramp & off-ramp crypto. This product may be too early for this wave of users but I'd like to have something that just feels like you've opened up a debit / savings account that you can give to your agent instead of providing access to your own details.
Thank you!
I'm building Soma. Soma is a platform where you can write & deploy AI agents or AI enabled workflows as code. Think n8n but without the messy visual notation, just regular code. We provide a single pane of glass to view, observe and monitor all your agents. And I'm looking for feedback ( http://trysoma.ai/ , old clickable prototype for the management UI https://wireframe.trysoma.ai/ ).
4 things that really stand out from our platform vs others:
1. We let you link multiple agents to a single inbox (Gmail, Slack etc.), and we intelligently route messages to a sub-agent you've deployed. Your end users see one agent (i.e. customer service agent, internal employee assistant or company book-keeper) and have a single unified entry point and UX for all linked workflows. Our in-house LLM routes requests or asks follow up questions until it knows what you want to do.
2.You can A/B test each deployments of the same agent. You can link custom metrics & dimensions to. So when you deploy an update to your customer refund agent, you can push custom data from Stripe or your customer satisfaction survey back into the platform and compare metrics between deployments over time.
3. It's regular code, with a few helper functions. Our SDK is based on https://restate.dev/ . TLDR; you can pause execution, replay failed workflows, and provide human-in-the-loop approval processes without custom business notations or DSLs. We've sprinkled some functionality on top of restate to enable common human in the loop patterns so you don't have to. We promote open standards (otel, A2A, MCP) to integrate with our platform and you can use whatever AI SDK (CrewAI, Vercel AI SDK) or none at all.
4. (A bit further down the line) We're building a loveable-style UI for non-technical power users to build out automations & workflows that your engineering team would be happy to take over, it's just plain old code :)
Which leads me to this: I want to invert the business model of a lot of these general AI agents (Manus, Claude, Motion etc.). What does that mean? I want to provide end users & engineers with the infrastructure building blocks that make these systems work.
Why? To make them truly personalised to you individually or as a business. I believe that Manus, Claude, etc. make assumptions about how you or your company works. And whilst every company has the same top-level process (e.g. end of month cash flow reconciliations [yawn]), they each have their own nuances and variations (e.g. company X wants to manually approve bank statement items over $1 000). I'd like to provide a set of tools where users can encode this level of nuance into each of their agents, deploy it onto our platform and orchestrate them together to make the overall user experience feel like a Manus or Claude.
We've started a private Alpha with 3 customers and we're looking for more individuals / companies to join our beta release as well as founding engineers & investors.
If this resonates with you, as:
1. An interesting technical challenge you'd like to find out more about, click here - https://form.typeform.com/to/zkAEOEod?typeform-source=www.tr...
2. An individual or company with a problem that this solves, click here - https://form.typeform.com/to/D8qL5efI?typeform-source=www.tr...
3. An investor interested in hearing more, click here - https://form.typeform.com/to/ffIyAVt1
Link to a clickable prototype of the management UI: https://wireframe.trysoma.ai/
Any feedback (good and bad) is welcome! I'm a one man shop for the moment, and the wireframes are a few months old as we're now building on top of the actual codebase. Apologies if they're a bit rough around the edges.
Thanks in advance, Daniel