Michael BrownJan 20
Federico Viticci 

An open-source AI agent running on my Mac mini server is the most fun and productive experience I've had with AI in a while.

Clawdbot by @steipete is a personal AI assistant that you can infinitely tweak. It runs in any messaging app (!), has access to skills, can use the shell, and – notably – it can improve itself because it's completely open to customization.

My story:

Clawdbot Showed Me What the Future of Personal AI Assistants Looks Like https://www.macstories.net/stories/clawdbot-showed-me-what-the-future-of-personal-ai-assistants-looks-like/

Jan 20 at 4:50pmWeb
Show threadDaron Brewood 🇬🇧Jan 20
@viticci @steipete Sounds awesome but rather expensive I imagine?
Show threadPeter SteinbergerJan 20
@tekguru @viticci Free and open source and the subs are cheap, most folks have the Anthropic one anyhow.
Show threadThomas TeichertJan 20
@steipete @tekguru @viticci so, this works with a Claude pro subscription? Or is a max subscription recommended?
Show threadPeter SteinbergerJan 20
@_ThomasTeichert @tekguru @viticci Works with pretty much any model and provider on the market. Currently Opus has the most character. OpenAI performs great is just more bland. MiniMax is extremely cheap and good ($2 fist month, then $10)
Show threadThomas TeichertJan 20
@steipete @tekguru @viticci amazing. When I read that viticci went through 180 million tokens, I was expecting something like $4,000 💀
Will have a look. Thanks Peter!
Show threadTobias HagemannJan 20
@_ThomasTeichert @steipete @tekguru @viticci You can actually burn through >2b tokens in a month with a $200 subscription, so… there‘s that. 😂
Show threadFederico Viticci Jan 20
@tobihagemann @_ThomasTeichert @steipete @tekguru I haven't plugged in my main Claude account yet for fear of getting banned but maybe I should try it 😜
Show threadDaron Brewood 🇬🇧Jan 20
@steipete @_ThomasTeichert @viticci It is on my list to look at in teh future, sounds vewt intriguing!
Show threadRaineer  🦆Jan 20

@viticci i vibed something like that just this weekend!

My use case is that I use DEVONthink (with MCP from Claude) as a back-end for all my project contexts and deliverables

I wanted a way to interact with that when I am away from the house (Claude desktop mcp only works on that system)

So now I have a remote ClaudeBot using the API to a local service on my Mac. Just in case I get the urge to vibe a new project def while standing in line somewhere 

Show threadQuentin Stafford-FraserJan 20
@viticci @steipete Sounds very scary to me.
It’s a nice idea but I’ve seen so many serious mistakes made by LLMs that I would never trust it with real data….
Show threadPeter SteinbergerJan 20
@quentinsf @viticci I run a version of this since May and nothing happened. The newer models made great progress on prompt injection. People in Discord tried really hard and clawd is laughing at them. There’s defo no 100% secure way, most folks there decided tho that benefits > risk.
Show threadQuentin Stafford-FraserJan 21
@steipete @viticci Glad it’s working ok - but I’m less worried about prompt injection than I am about the models just making mistakes!
Show threadPeter SteinbergerJan 21
@quentinsf @viticci I run this since May and the worst thing that once happened was it corrupting a git repo when I asked for a complex history rewrite.
Show threadWarren GallagherJan 20
@viticci What Mac Mini configuration are you running? The Claude model is remote so perhaps the specs are unimportant?
Show threadShaunJan 20
@viticci this sentence low key terrifies me as a developer (I don’t disagree with you, it’s just also VERY SCARY)
Show threadManuel T. GomesJan 27
@viticci @steipete was there a name change?
Show threadFederico Viticci Jan 27
@manueltgomes @steipete Yes, and updated my post at the top.