visionik e/acc
1.5K posts

visionik e/acc
@visionik
Dad/Husband/GliderPilot/SpaceNerd/f≡ounder,i≡nvest;6502→.PAS→C→JS→⨂(CAC→LTV→ROI+NPS)→⎋³ᶠ→S.A.F.E→.PY→LP→SPV→⎋³ⁱ→.GO→.TS→e/acc→AI→Inc∈’26=+3 @openclaw maintainer
Orlando, FL, USA Katılım Kasım 2007
1.5K Takip Edilen1.4K Takipçiler

deft.co day 8.0.0 Changelog
(Note: author was in California on another project for ~a week)
Ops:
- many accounts created
- many computers ordered
- many shelves built
Universe:
- Moved repos to github.com/deftai organization
- Deft Directive moved to trunks
- Deft ████████████ added standards-based █████
- Deft Brief Studio v0.0.1
- Deft Console plugin brainstorming
- Deft Bridge plugin implementations
Office:
Desk deep-dive...
1. 5G backup internet
2. USB-C PD power
3. New monitor running... something
4. Nice desk lamp
5. Apple HomePod (1 of 3)
6. Remote for fancy Dyson fan
7. Temporary OPAL replacements
8. New MacBook Pro M5 Max
9. Orange Stanley quencher
10. Orange Beefy King beanie

English

@cjpedregal CLI. Please. It's the right way today and in the future.
English

There are some tweets out there saying that Granola is trying to lock down access to your data.
Tldr; we are actually trying to become more open, not closed. We’re launching a public API next week to complement our MCP. Read on for context.
A couple months ago, we noticed that some folks had reversed engineered our local cache so they could access their meeting data.
Our cache was not built for this (it can change at any point), so we launched our MCP to serve this need. The MCP gives full access to your notes and transcripts (all time for paid users, time restricted for free users). MCP usage has exploded since launch, so we felt good about it.
A week ago, we updated how we store data in our cache and broke the workarounds. This is on us. Stupidly, we thought we had solved these use cases well enough with our MCP.
We’ve now learned that while MCPs are great for connecting to tools like Claude or chatGPT, they don’t meet your needs for agents running locally or for data export / pipeline work.
So we’re going to fix this for you ASAP. First, we’ll launch a public API next week to make it easier for you to pull your data.
Second, we’ll figure out how to make Granola work better for agents running locally. Whether that’s expanding our MCP, launching a CLI, a local API, etc. The industry is moving quickly here, so we’d appreciate your suggestions.
We want Granola data to be accessible and useful wherever you need it. Stay tuned.
English

@0thernet @zocomputer Practically every idea is one degree from another idea.
Constructive criticism: I understand your arc/frustration but it's feeling a bit cringe to me.
English

hilarious how many people are copying @zocomputer
> June 20, 2025 – Zo beta
> Nov 19, 2025 – Zo launch
> 6 days later – first OpenClaw commit
> Feb 2026 – Every YC company pivots to cloning Zo
MuleRun@mulerun_ai
Introducing MuleRun 2.0. Your personal AI, act before you ask. It learns your habits, anticipates your needs, and works while you sleep — running 24/7 on your Personal Computer assigned to you alone. No complex setup. Just talk to it.
English

@cjpedregal @appenz @meetgranola MCP is not only an awful strategy; it's a dying strategy.
blog.visionik.com/mcp-is-evil-2/
Have you not noticed how many products have added a comprehensive CLI and are moving away from MCP?
English

@appenz @meetgranola Hey Guido, we're not going closed. Our MCP lets you access everything. If you have feedback on why MCP isn't helpful, or what else you'd prefer, let me know plz
English

Sorry to see Granola @meetgranola going closed. They encrypted their local db, no local and no cloud API. In a world where notes are managed by agents, the app now has zero value.
Any recommendations for good alternatives? What are you switching to?
English

@CardilloSamuel I couldn't disagree with you more about @steipete.
I couldn't imagine it will make any difference to you.
English

btw i am switching fully from openclaw to hermes agent, mainly because i dislike peter but also because hermes is ten thousand times better so far.
i've met tons of very egoistical persons in my life but peter is like, highest level. all of his public interactions just feel like he think he's some kind of above-all humans and it piss me off. compare that to the people who maintain hermes agent and how close to the community they are.
big difference.
English

@Bob_Wachter My em-dash's are in print and original Mac Pagemaker files from 1988 — I want grandfathered in too, please.
English

@uzakyolkaptani @spaceship I bought it. The Deft team and I thank you!
English

Honestly, this update made my day! Sold a domain for $12,500 (full commission paid by the buyer). Held it since Oct 2024. BIN was $19,888.
Congratulations to the buyer and thanks to @Spaceship!

Cemal Coban@uzakyolkaptani
@namecheapceo123 Interesting update.👍🏻
English

This is a worthy cause that is near and dear to my heart (literally and figuratively). I invite you all to think about how you can leverage it in your own work. nepotism.org
(designed, built, tested, and deployed by deft.co in under 20 minutes)
English

5/ I've been b*tching about this for years. This AM I bought boardscale.com & pcbscale.com domains. I found a 125k sq ft industrial building @ an ideal location. And "we" got our first co-founder on board.
We will slay PCB's in the USA. Watch this space.
English

Hi Adam, YW and thanks for sharing your thoughts.
- I agree that code mode solves all? most? of the context window bloat, but I still contend CLI's are the better pattern. Code mode is essentially just dynamically creating small CLI-like tools to use the MCP. Why not just ship CLI's?
- Discovery: I agree, and its why I created github.com/visionik/dashd… to have a good pattern for both "ability" discovery and "access method" discovery.
- MCP leaves security entirely in the hands of the MCP authors. It creates a new kind of "thing" for which we have to create new security strategies. CLI's have been around for 60+ years. We have many security patterns which can be applied to them. access rights, containers, sandboxes, wrapping them in shell scripts of other CLI's, etc. Yes, you can also do most of this in MCP; but all of this gets back to my OG point: you *can* do everything MCP does in CLI's. Why did we need to invent a new thing?
Enjoying the discourse, -V
English

Thanks for sharing this. I have a few thoughts, and I’m curious what you think.
- “Context window” bloat isn’t an MCP issue. It comes from function calling itself. It’s been addressed by Code Mode, where the agent gets a code execution tool and can explore available schemas when needed, without loading them upfront. Code execution also allows the agent to compose tools or “pipe” them.
- Both Code Mode and the CLI have the same issue: the agent has to somehow discover the available interface, because it won’t know it upfront unless the CLI is well-known for LLMs, like `github-cli`, for example. This requires the agent to explore using `--help` or schemas, which can easily lead to loading tons of information into the context that the agent doesn’t need. Moreover, this way, an agent may never find what it actually needs because they are like “unknown knowns” to them.
- Both “Code Mode / Code Execution” and the CLI (which requires the agent to have access to a terminal) come with the same power and the same flaws. The power comes from the fact that LLMs are great at using them. The flaws come from the fact that you have near-zero control over what the agent will do when it is connected to the Internet. In fact, Code Mode with MCP is a little more secure and steerable because, at least to some extent, you can limit what the agent will be able to do, but it often does not help much anyway.
- Neither MCP nor the CLI is secure in an agent’s hands because prompt injection is still an open problem. You may receive a malformed email, or an agent may read a website that causes it to expose your data by pulling it from other tools and then sending it via a simple API call. It doesn’t matter whether you use MCP or the CLI; So, “Security Nightmares,” “Tool Poisoning,” and “Cross-Context Data Leaks,” which you mentioned in the article, are common to both scenarios.
And to be clear - I'm not here to convince you wether MCP is better or makes sense. In fact MCPs today are pretty much all about exposing tools, which can be done in many other ways.
I'm just curious what you think about this.
English

Is it the death of MCP? CLI is all you need.
Just today @firecrawl and @p0 announced a CLI
MCPs are a solution for humans pretending to build for agents. Agents are processes. Processes talk to the world through CLIs and APIs. Always have.
I expect most companies to quietly drop MCP and go back to basics.
English
visionik e/acc retweetledi

Deft AI day 2.0.0 Changelog
Office:
- $1M in monitors! (joke)
Universe:
- Deft Directive v0.6.0 (deft.md)
> Slash cmds!
> YOLO (harder)!
> Unity!
- Deft Evolution v0.0.12
> 12 / 65 phases done
> 98.08% unit test coverage,
> Built by the "sales guy".

English

Deft AI log 1.0.0
Office:
- Unifi v1.0.0
- Back v0.0.2
- Front v0.0.1
Universe:
- Directive v0.5.2 (deft.md)
- Evolution v0.0.1
- Captain v0.2.0
- BRIEF v0.6.0 (vbrief.org)
- Communicator v0.5.0
- Console v0.6.8
- Bridge v0.2.0


English

@Il_Pastore_TCMC @digitalix Well... the worst kind of racks are the ones with pre-threaded holes. You strip one threaded hole in the wrong place and it's game over man.
One other guy mentioned it; forget rack nuts or even rack studs. This is the way:
shop-us.patchbox.com/products/dev-m…
English

@digitalix May I suggest:
shop-us.patchbox.com/products/dev-m…
shop-us.patchbox.com/products/patch…
Also this is super handy not just for installing switches etc., but it can also be reversed and used as a laptop stand while you are at the rack.
shop-us.patchbox.com/products/setup…
English





