@kepano@_thinking_reed Don’t know what he meant but this CLI is only available if you have the Obsidian Desktop. Not with Obsidian headless for our remote agents to use it
Unpopular Opinion: We aren't building the future 10x faster with AI. We are just generating legacy code 10x faster.
Everyone is currently bragging about developer velocity. "I built this entire backend in a weekend!" "AI wrote 80% of my codebase!"
But here is the reality check we are ignoring: Code is a liability, not an asset.
If an AI tool spits out 1,000 lines of functional boilerplate in five seconds, that is still 1,000 lines that a human being has to read, review, secure, and maintain when the dependencies inevitably break next year.
We are treating code generation like a pure productivity win, but we are optimizing for the wrong metric. The bottleneck in software engineering was never how fast we could type. The bottleneck has always been comprehension, architecture, and maintenance.
If we don't shift our focus from "generation speed" to "architectural sanity," the tech debt of the next five years is going to be an absolute, unmaintainable nightmare.
@dsp_ Funnily enough, the MCP spec defines a server definition that could be loaded in the context and have a similar progressive disclosure as Skills.
But MCP clients chose to load all the tools instead 🤷♂️
@ErikVoorhees@AskVenice Not storing is a good thing, but we’ve been asking about who, where and how the AI inference is done.
Your doc says “a network of decentralised providers”, with no mention of guaranteed privacy from these providers.
And yesterday you contradicted it saying you run it yourself
@ErikVoorhees@milianstx How is it not documented anywhere?
You privacy page mentions a “decentralised network of providers” and now you say you operate it yourself.
Is it TEE? SSL termination in the TEE? Any attestation?
Otherwise it’s just “trust me bro”
https://t.co/CblX7ICRef
@ErikVoorhees@kenziyuliu@openclaw@steipete Thanks for the reply. You doc still mentions you use decentralized GPU providers, so my question still applies
https://t.co/ELu04xGbrR
@ErikVoorhees@kenziyuliu@openclaw@steipete For the open source models that you run, you actually use Akash on decentralised GPUs.
What are the guarantees there is no eavesdropping by the actual computer providers there?
@ErikVoorhees Until you provide the model with some personal data that likely contains PII.
Unless the open models run in TEE can’t be seen by the inference providers?
@chrysb@openclaw If only the MCP spec would recommend that clients are only putting the server description into the context and lazily (instead of eagerly) loading the tools when needed, then we would a very similar token efficiency as the skills.
Would not even need “tool search tool”
@ErikVoorhees To my understanding, your inference is done on Akash by a decentralised network of GPU. Where’s the guarantee the GPU providers don’t look into the data? Are they running in TEEs?
@Kautukkundan Good insight on the MCP ecosystem. But, MCP is just a spec for the connectivity.
If consumer apps don't allow technical integration it's not MCP fault.
The "Traditional MCP approach" you also describe is not related to MCP but to the AI apps/agents using it
@udaysy@RhysSullivan It’s because the MCP specification “enforces” both the server and the client to support the different elements for a smooth OAuth flow
@championswimmer@openclaw Why the need to distinguish a token for LLM vs for a “human”?
It represents the ability to do what the human initially consented for