Benchmarks

The thesis

Similarity is not relevance.

A longer context window does not make an agent stateful. Real continuity comes from structure. HydraDB stores knowledge as a versioned graph that preserves every state transition, enriches each fact so it stands on its own, and fuses semantic, lexical, and relational signals at recall time.

Beyond Flat Embeddings for Production AI Agents

How HydraDB's context graph outperforms flat vector retrieval across eight public benchmarks: enterprise knowledge, finance, code and long-term memory.

Retrieving the Right Memory Across Four Years of Personal Context

HydraDB reaches 0.918 recall@10 over 11,034 personal photos, videos and emails on ATM-Bench, second on the public leaderboard and first among systems using the benchmark's own captions.

Retrieving Correct Context Among Adversarial Look-Alikes

HydraDB reaches 0.917 recall@k and 0.776 nDCG@k on KyroBench's stale, cross-tenant and poisoned look-alikes, 0.338 nDCG above the best baseline.

Retrieval Over Two Centuries of Federal Financial Statements

HydraDB finds the right document among 1,435 near-identical U.S. government financial tables from 1793 to 2024, reaching 81.1% recall@10 on the cited set.

Building Coding Harness on HydraDB

A graph-backed documentation harness scores 90.73 on nlohmann/json, compared with 66.06 for DeepWiki and 61.28 for CodeWiki.

Retrieval Over Company-Internal Knowledge for AI Agents

HydraDB retrieves connected, current context across a synthetic corpus of more than half a million documents from nine fragmented enterprise systems.

overall · 10 dims

Long-Term Memory for AI Agents at Million-Token Scale

Why million-token context is not the same as useful memory, and how HydraDB reaches 82% overall on BEAM 1M.

Remembering What Is Still True Across Long Conversations

HydraDB reaches 90.79% on LongMemEval-s, 5.0 points above the strongest reported system, with 97.43% on knowledge updates and 85.80% even with GPT-5 Mini.

The Context Substrate for Financial AI

How HydraDB turns dense filings and enterprise data into relevant, connected, current, traceable, and compact context for financial AI.