HOP - the code is much more reliable than at the start of the year. SafeFlow-ECS is at the heart and some improvements but no major refactoring. Most of the effort has gone into adding coherence and besearch cycles (still a WIP) and adding beebee agent, direct and other agents via gpt4all library. We have the opportunity to show two key aspects of decentralized machine learning:
@teach where a network of peers can 'fine tune' a model to use the BentoBoxDS and form HOP
DML via a besearch cycles - shows how many science experiments can be evolved and how a network of peers can evolve and achieve coherence and eventually consilience on the science. These besearch cycles will be manged by TINY LLM agents.
Model: HOP has a new model folder where the SafeFlow-ECS compute engine. The opens up the use of cryptographically signed code. A new compute engine for SafeFlow was written to use this setup.
Peer to Peer Connection: This has been upgraded and more reliable on connecting peer for the first time.
Model: HOP has a new model folder where the SafeFlow-ECS compute engine. The opens up the use of cryptographically signed code. A new compute engine for SafeFlow was written to use this setup.
Peer to Peer Connection: This has been upgraded and more reliable on connecting peer for the first time.
BentoBoxDS - overall the peer experience has been improved. This coming from the use of coding agents to write e2e tests and css code. BentoBoxDS has add beserach (will be the new default peer experience) and rebuild chat interface using beebee agent. The Library has been neglected and testing of besearch and peer to peer machine learning notifications is the task at hand.
OpenHands open source coding agent. Half the code, especially BentoBoxDS was achieved in first half of year with goal to build 90% by close of 2025. BentoBoxDS web, desktop and mobile will be rewritten using bare.js instead of node JS. This is a major refactor and this scale of change is only possible because of these new coding agents tools.
TINY devices and data - choice to focus on three sources, Polar, Withings and Blood biomarkers around healthCues on bentoboxDS so show case. Export data, API data and white labeling storing direct to peers personal data store.
Criticism and looking ahead - the product ambition continue, scope a capability expands again in 2025 and nothing quite gets solidly and rigorously connected all together allowing a peer to use the project, device to seeing a chart of bentoboxDS. This has to stop and should as the conceptual thinking is now all expressed in code. Coding agents now equaling up the ambition with delivering a working product. Having said that we need to focus on complete peer experience and automate distribution i.e. register app in Window, App store and mobile offerings.
2026 is all about demonstrating peer teaching their tiny agents to learn and for that learning to coherence across all the peers on the network.
OpenHands open source coding agent. Half the code, especially BentoBoxDS was achieved in first half of year with goal to build 90% by close of 2025. BentoBoxDS web, desktop and mobile will be rewritten using bare.js instead of node JS. This is a major refactor and this scale of change is only possible because of these new coding agents tools.
TINY devices and data - choice to focus on three sources, Polar, Withings and Blood biomarkers around healthCues on bentoboxDS so show case. Export data, API data and white labeling storing direct to peers personal data store.
Criticism and looking ahead - the product ambition continue, scope a capability expands again in 2025 and nothing quite gets solidly and rigorously connected all together allowing a peer to use the project, device to seeing a chart of bentoboxDS. This has to stop and should as the conceptual thinking is now all expressed in code. Coding agents now equaling up the ambition with delivering a working product. Having said that we need to focus on complete peer experience and automate distribution i.e. register app in Window, App store and mobile offerings.
2026 is all about demonstrating peer teaching their tiny agents to learn and for that learning to coherence across all the peers on the network.
————————————————————————————-
week 27 October 2025
Four development strands that need woven together
HOP update places locations, how to handle besearch flows, DML and train agent for beebee, selection of AI agents
BentoBoxDS (receive input places from HOP, update modules to update data (and automate), besearch cycles peer and peer to peer)
Tiny device dataflow (polar JSON parser API partnership call, WIthings csv files praser API integration, partnerhships)
RGB/payment infrastructure
Tiny device dataflow (polar JSON parser API partnership call, WIthings csv files praser API integration, partnerhships)
RGB/payment infrastructure
Upload Withings exported csv files. Weight file. Update HOP SafeFlow-ECS for
If files Hyperdrive and Hyperbee. make path
nxpID file type filename (should name be hashed? or keep english?)
How to handle file updates?
Keep format for API calls?
How about
—————————————————————