Skip to content

armgit / ruflo

Public Imported from GitHub
0

🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated

darwin-plugins iter 2 β€” ruflo-agentdb: replace unverified 150x-12,500x HNSW claims with measured ~1.9x-4.7x

ruv6ae2746ee47d
diff --git a/plugins/ruflo-agentdb/skills/vector-search/SKILL.md b/plugins/ruflo-agentdb/skills/vector-search/SKILL.md
index 386818bd75..e12f2ffd4d 100644
--- a/plugins/ruflo-agentdb/skills/vector-search/SKILL.md
+++ b/plugins/ruflo-agentdb/skills/vector-search/SKILL.md
@@ -11,7 +11,7 @@ Two distinct vector-search paths live in this plugin. Pick the right one β€” the
| Path | Tool family | Backing | Capacity | Latency |
|------|-------------|---------|----------|---------|
-| **Large-scale corpus** | `embeddings_*` | `@claude-flow/memory` HNSW (Rust/Native) | up to millions of vectors | 150×–12,500Γ— faster than brute-force, depending on N and parameters |
+| **Large-scale corpus** | `embeddings_*` | `@claude-flow/memory` HNSW (Rust/Native) | up to millions of vectors | ~1.9Γ— at N=20k, ~3.2×–4.7Γ— at N=5k vs brute-force (measured; recall@10 β‰ˆ 0.99). ANN wins above the crossover |
| **Hot-path router** | `ruvllm_hnsw_*` | WASM-backed router (v2.0.1) | **~11 patterns max** (`ruvllm-tools.ts:58`) | sub-ms; designed for high-priority routing, not corpus search |
The "12,500Γ—" headline applies to the large-scale `embeddings_search` path. The WASM router is **not** that path.
@@ -85,10 +85,13 @@ npx @claude-flow/cli@latest memory search --query "your query"
## Performance
-| Method | Speed |
-|--------|-------|
+Measured numbers (source: `scripts/benchmark-intelligence.mjs`, ruvector NAPI backend; recall@10 β‰ˆ 0.99). The older "150×–12,500Γ—" figures were brute-force-fallback artifacts and have been retired β€” see project CLAUDE.md "V3 Performance Targets".
+
+| Method | Measured speedup vs brute-force |
+|--------|---------------------------------|
| Brute-force scan | Baseline |
-| HNSW (n=500, balanced) | ~150Γ— faster |
-| HNSW (n=10,000, balanced) | ~12,500Γ— faster |
-| RaBitQ + rerank (n=10,000) | ~12,500Γ— search speed at 32Γ— lower memory |
+| HNSW (N=5,000) | ~3.2×–4.7Γ— faster |
+| HNSW (N=20,000) | ~1.9Γ— faster |
+| HNSW (below crossover, small N) | ties/loses vs brute-force |
+| RaBitQ quantization | 32Γ— memory reduction; 0.60 ms/query at Nβ‰ˆ14.7k |
| `ruvllm_hnsw_route` (n≀11) | sub-ms per route, fixed cost |