A protocol and engine by BEBEL
Built for the speed of the next iteration.
Lightspeed is the speed, latency, and parallel-iteration protocol we are building at BEBEL, and the family of engineering and architectural systems built on it. It is designed to take work apart and rebuild it quickly and cheaply, many attempts at a time, while using memory and hardware efficiently. Our Personal OS is the first product built on it.
What it is
A protocol first, then an engine, then products.
The protocol is a common language for people, agents, and services. Every message travels as a signed, content-addressed event over plain HTTPS, so any language on any host can take part and every message can be checked later.
The engine runs each obligation, whether a task, a question, an alert, a decision, or a release, as a durable workflow with an owner, a deadline, and a receipt that can be read back. Work can run in parallel, fail cheaply, and be rebuilt from its record instead of from memory.
Lightspeed is in active development. The protocol is a working draft, and nothing is publicly released yet.
The working thesis
Speed is designed, not hoped for.
- 01Treat latency as a feature
Measure the time from question to useful answer and design every layer to shorten it.
- 02Iterate in parallel
Run many attempts at once and keep the one that holds up, instead of waiting on a single path.
- 03Tear down and rebuild cheaply
Make every part replaceable and every run reproducible, so changing course costs little.
- 04Spend memory and hardware deliberately
Choose the smallest model, machine, and storage that does the job well, and know what each run cost.
- 05Leave a receipt for every claim
Record what was done, by whom, and with what result, so the next iteration starts from evidence.
- 06Outlast the vendor
Keep the protocol independent of any one AI provider, cloud, interface, or machine.
Why we chose the name
Context at the speed of the question.
The system should remember. The person should remain free to change.
A question arrives in a moment. The history and the tools needed to answer it usually do not: they are scattered across systems, inboxes, devices, and old conversations, and they are rebuilt from scratch each time.
We chose the name Lightspeed because the protocol and engine are being built to close that gap: to bring the right record and the right capacity to the question quickly, whichever device or AI provider is asking, and to keep what was known, what was chosen, and what changed since.
AI can help search that record, question a first answer, compare interpretations, and carry bounded work forward. Neither a fluent answer nor a complete archive gets to decide what a life means. A system can offer another view. The person lives with the decision.
Related research
Two problems the product has to solve.
What should a system do when evidence is incomplete? What can automated review miss?