Talk
Beyond the Clouds. Notes from BKK Hack
We presented Beyond the Clouds at BKK Hack. Cloud chat made strong models easy to reach. It also moved memory, tools, and work onto infrastructure you do not control. The talk is about what now runs on your computer, why persistent workers change the risk, and where Velociti sits.
Chat became ongoing work
A chatbot waits for a prompt and forgets the job when the tab closes. Persistent workers such as OpenClaw and Hermes keep goals, memory, and tools. They can run on a schedule, act across sessions, and come back with a result.
The more an assistant remembers, and the more systems it can reach, the more the data boundary matters. Prompts, context, responses, and thinking stay on your computer only if the workspace is built that way.
Open-weight models made local possible
DeepSeek, Qwen, Kimi, GLM, and others released capable weights that independent users can host. Open-weight is not the same as open-source. Weights can be public while data, training code, and licence still have limits.
That shift is why a local-first AI workspace is practical now. A desktop with a current GPU and enough system RAM can run a strong local model for private drafting, extraction, and ongoing agent work.
The missing middle
Hosted labs sell strong models with a recurring bill and an external data boundary. Hobby stacks maximize control and cost a lot of time to keep alive.
Most teams sit in between. They need ownership, governance, and a record of what the assistant saw and did, inside work they already run. That gap is the product.
What Velociti is for
Velociti is a local-first AI workspace. Choose a local model or bring your own key. The project, files, and approved memory stay on your machine. A supported provider enters only when you send a turn out.
For companies, Velociti Enterprise is the control plane: who can act, what tools are in scope, and a receipt of what happened. Intelligence can come from anywhere. Authority stays with you.
The BKK Hack slides are on this page as a PDF download. Then read how Velociti works.
Questions
- What is in the download?
The PDF slidedeck we used at BKK Hack, Beyond the Clouds. It covers the timeline from chat to persistent workers, local hardware, and Velociti as the missing middle.
- Do I need a cluster to run local AI?
No. Fit follows measured headroom on your machine. See run local models on your computer.
- Is this a transcript of the talk?
No. This is a short written version of the argument. The slides are the deck we presented.