# Retrieval and memory for agent teams

- Page: https://scanovich.ai/research/retrieval-and-memory/
- Language: en
- Updated: 2026-11-11

- Status: active

**Question.** How should a team of AI agents share memory so that a wrong note cannot outrank a reviewed fact?

## Why it is uncertain

Several agents write notes into one memory. Some notes are reviewed, most are not. A retrieval system ranks by relevance, and a confident false note is very relevant. Nobody had measured, on our own setup, how often the false note wins.

## Method

- Hybrid retrieval that runs locally: dense and sparse vectors, then a reranker.
- A benchmark with questions in Russian, English and mixed language.
- A poisoning test: false notes are planted against known facts, and the ranking is measured with and without trust tiers.
- Two tiers: curated files and unreviewed notes. The ranking can prefer the first or use it alone.
- The whole harness is public, so the numbers can be rerun.

## What we measure

- Whether the right passage is ranked first, and whether it is in the top five.
- How often a planted false fact is ranked first.
- Latency per query.

## Results so far

On the public synthetic set of 96 passages and 24 questions, search with reranking puts the right passage first for 92% of questions. Plain text search manages 38%.

Without trust tiers, a planted false note that contradicts a reviewed fact is ranked first in 6 of 6 cases. With reviewed files preferred, in none.

One limit is measured too. Tiers cannot tell two unreviewed notes apart: when the truth itself is only an unreviewed note, the false one can still win.

## Next

Rerun the reranker comparison on real trade documents instead of the demo set.

## Projects on this line

- [Services-BGE](https://scanovich.ai/projects/services-bge.md) — released, open source · MIT. Local embedding and reranking services for hybrid search, and a shared, measured memory for AI agents.

## 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