[biddit]

I have a bespoke local agent that I built over the last year, similar in facilities to Moltbot, but more deterministic code.

Running it this kind of agent in the cloud certainly has upsides, but also:

- All home/local integrations are gone.

- Data needs to be stored in the cloud.

No thanks.

[aaravchen]

This is exactly the issue. Even if you ignore the privacy concerns, the reason ClawdBot/Moltbot/OpenClaude got so popular is that everything was actually run locally. The early adopters where people on locked down corporate networks where almost everything they need to interact with is in the category of "a local printer" (possibly a networked one).

Cloudflare simply cannot access anything most users will want to access. If it's not run locally, it simply won't work for most users.

Piled on top is the obvious data privacy issue. Most notably the credential privacy, but also the non-credential privacy and data collection. Hard pass from me until there's a solution that covers all of these, including personal data privacy (and a "privacy policy" is no privacy at all).

[mitchitized]

This is ultimately the first question I have whenever someone tells me about a bouncing new AI shiny... "Where does my data go?" Because if it does not stay on my machine, hard pass.

[KurSix]

There's a hidden trade-off here: Latency vs Privacy

A local agent has zero ping to your smart home and files, but high latency to the outside world (especially with bad upload speeds). A cloud agent (Cloudflare) has a fat pipe to APIs (OpenAI/Anthropic) and the web, but can't see your local printer.

The ideal future architecture is hybrid. A dumb local executor running commands from a smart cloud brain via a secure tunnel (like Cloudflare Tunnel). Running the agent's brain locally is a bottleneck unless you're running Llama 3 locally

[halfcat]

What kind of hardware do you need, and how is it compared to the cloud agents?

[theshrike79]

I've been thinking of a similar thing, I just need a local model with consistent tool calling performance.

Most of my crap could just be tools and a mid-level language model interpreting the results and deciding whether to act on them.