I would like to share my work on a token saving protocol I call Slipstream (StreamLined Inter-agent Protocol… stream) which I maintain an open source implementation of on Pypi and Github called slipcore. Using semantic anchors and a universal concept reference (a 4d semantic manifold similar derived from how embeddings worked but standardized to work across models) we are able to achieve 82% token savings in swarms using models fine-tuned to speak Slipstream natively. (available here on hf)
More information in the article and space - a fascinating ai safety case study came out of the development of this protocol called “The Manifold Incident” - you can find information about that in my profile as it has a space explaining the paper. I self publish on Zenodo because I am an independent research and have no endorser on arXiv. If you think my work has merit, please contact me for my endorsement code. I particularly want to share my recent paper on the Manifold Incident (Living of the Land / Model Organism) case study.
Thank you for reading.