# Services-BGE

- Page: https://scanovich.ai/projects/services-bge/
- Language: en
- Updated: 2026-11-11

- Status: released · v0.3.0
- Code: open source · MIT
- Repository: https://github.com/FUYOH666/Services-BGE
- Research line: [Retrieval and memory for agent teams](https://scanovich.ai/research/retrieval-and-memory.md)

Local embedding and reranking services for hybrid search, and a shared, measured memory for AI agents.

Search that runs on one machine and can be measured.

## What works today

- Dense, sparse and late-interaction vectors from one model, then a reranker, behind a standard interface.
- Models can be swapped by profile and compared with the built-in evaluation harness.
- A memory server for agents: remember, search, and report what is indexed. Several different agents use the same memory.
- Trust tiers: reviewed files can outrank unreviewed notes, or be the only source.

## Why it is in the lab

It is the instrument for the retrieval and memory research, and it serves embeddings to the document projects.

## Evidence

- **92%** — of questions get the right passage ranked first by search with reranking. Plain text search manages 38%. Measured 2026-11-11. Source: https://github.com/FUYOH666/Services-BGE/blob/main/docs/MEMORY-BENCH.md
- **6 of 6** — times a planted false note wins without trust tiers. With reviewed files preferred, it never does. Measured 2026-11-11. Source: https://github.com/FUYOH666/Services-BGE/blob/main/docs/MEMORY-BENCH.md
- **0.861** — recall at rank one with the better of two rerankers on the Russian demo set. The other reaches 0.750. Measured 2026-11-11. Source: https://github.com/FUYOH666/Services-BGE/blob/main/eval/results/COMPARISON.md

## How to engage

An audit of your existing search or assistant, measured on your own questions with this harness. Private document search installed inside your network.

[Contact](https://scanovich.ai/contact/) · private@scanovich.ai