| _version_ | 1866901848993038336 |
|---|---|
| author | Diogenes |
| author_facet | Diogenes |
| contents | <p>This technical note proposes the <em>Ephemeral Context Federation Layer</em> (ECFL): a privacy-first way to give large language models apparent “memory” across conversations, without persistent storage, user profiling, or long-term logging.</p> <p>Instead of saving embeddings or chat histories in a database, ECFL builds a temporary, in-memory vector index by federating semantically relevant snippets from a user’s prior sessions. The index lives only for the duration of a single answer: it is created on demand, queried once to assemble contextual prompts, and then securely zeroized. No long-lived memory, no cross-session tracking — just <em>recall without retention</em>.</p> <p>The architecture combines dense retrieval, lexical search, and diversity-aware reranking (MMR), wrapped in a strict token budget and guarded by privacy and safety policies. ECFL can be deployed in server-side RAM, on-device (via WebGPU), or inside trusted execution environments (e.g. SGX/SEV) with remote attestation.</p> <p>The goal is simple: help LLMs remember what matters in a conversation.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17798327 |
| institution | Zenodo |
| language | |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Ephemeral Context Federation:\\ Giving LLMs Memory Without Becoming Stalkers Diogenes <p>This technical note proposes the <em>Ephemeral Context Federation Layer</em> (ECFL): a privacy-first way to give large language models apparent “memory” across conversations, without persistent storage, user profiling, or long-term logging.</p> <p>Instead of saving embeddings or chat histories in a database, ECFL builds a temporary, in-memory vector index by federating semantically relevant snippets from a user’s prior sessions. The index lives only for the duration of a single answer: it is created on demand, queried once to assemble contextual prompts, and then securely zeroized. No long-lived memory, no cross-session tracking — just <em>recall without retention</em>.</p> <p>The architecture combines dense retrieval, lexical search, and diversity-aware reranking (MMR), wrapped in a strict token budget and guarded by privacy and safety policies. ECFL can be deployed in server-side RAM, on-device (via WebGPU), or inside trusted execution environments (e.g. SGX/SEV) with remote attestation.</p> <p>The goal is simple: help LLMs remember what matters in a conversation.</p> |
| title | Ephemeral Context Federation:\\ Giving LLMs Memory Without Becoming Stalkers |
| url | https://doi.org/10.5281/zenodo.17798327 |