If I want to give an agent autonomy to make trades on my behalf — today, without writing any code — what’s the best way to do it?

Haven’t been keeping up with all of the developments but I’d like to eventually outsource trading decisions for most of my (not so large) crypto portfolio.

Founder @ robo.tech — AI agents for biotech

Previously founder @withBackdrop (acquired by Seed Club)

what types of trading decisions would you like to offload? do you have a particular strategy that you want to follow (e.g. momentum trading), or do you basically want to just watch what you have and maybe have it tell you when it does something "weird"?

Whipped up this POC for a trading agent strategy a while back when I was going to build it. 

Needed to prioritize a few other things so open sourcing this — if someone builds it, I'll put in the first 10k https://docs.google.com/document/d/1JQh5iPGoMLtAz9wS-HJigs1B52c4_pHanz3oa-_9cNo/edit?tab=t.0

Basically, this https://warpcast.com/joey/0xab3c3089

I want it to make trades following a specific strategy. Could be fairly tightly scoped to start, and then over time ideally it becomes more iterative/intelligent using NLP and other data sources.

We built something called Silverback that can implement those tightly scoped strategies extremely well, you write them as Python scripts (bots) that can listen and respond to blockchain events in real-time, including sending transactions for trades, etc.

We have a "bot cluster" product that just came out last week, basically you can run tens to hundreds of these strategies on multiple netwotks and manage them very easily

I've been thinking about the agent piece, basically wrap the bot cluster control interface with a fairly typical LLM chat agent that knows how to search for some bots to run based on metadata of their container images describing what they do, and then knows how to use the bot cluster as a tool to install and start them running, occasionally check their health, etc.

The bots themselves have really strong foundations with our devtooling, which people use for contract development and chain action all the time. This minimizes the surface that the LLM would get things wrong

What do you think?