@neruva/mcp 0.27.1 → 0.27.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +5 -5
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -8,13 +8,13 @@ Drop into Claude Code / Cursor / Codex / Gemini CLI / Goose in one line. **Free
|
|
|
8
8
|
|
|
9
9
|
> **For Claude Code users:** see [neruva.io/claude-code](https://neruva.io/claude-code) for the 30-second install + first-queries to try.
|
|
10
10
|
|
|
11
|
-
##
|
|
11
|
+
## Benchmarks
|
|
12
12
|
|
|
13
|
-
|
|
|
13
|
+
| Test | Score | What it means |
|
|
14
14
|
|---|---|---|
|
|
15
|
-
| **
|
|
16
|
-
| **
|
|
17
|
-
| **
|
|
15
|
+
| **Does the agent learn from mistakes?** | **+34 points** | Same Claude Haiku model, no retraining. With Neruva running the agent climbs from 84% to 93% across 2000 tasks. Without Neruva it stays flat at 59%. Three independent runs. |
|
|
16
|
+
| **Memory QA across long histories** (LongMemEval) | **93.3%** | top-4 globally, +22pp vs Zep |
|
|
17
|
+
| **Compliance + audit determinism** (DFAH) | **100% / 88%** | first system to do both at once, 2.75× prior best |
|
|
18
18
|
| **Latency** (p95 cache-hit) | **80ms** | 2.5× faster than Mem0 |
|
|
19
19
|
|
|
20
20
|
Full breakdown: [neruva.io/benchmarks](https://neruva.io/benchmarks). **The only memory stack hitting world-class scores on memory, reasoning, AND audit determinism — simultaneously.**
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@neruva/mcp",
|
|
3
|
-
"version": "0.27.
|
|
3
|
+
"version": "0.27.2",
|
|
4
4
|
"description": "MCP server for Neruva -- reliability infrastructure for production AI agents. One install, agent stops forgetting / drifting / repeating mistakes, every failure replayable bit-for-bit. v0.27 ships the qbound KG engine -- pick engine='qbound' on hd_kg_add_facts when document-stated facts must override LLM training-time priors. Now 6 KG engines (hadamard, opb, qbound, multishard, quorum, feature_bundle). v0.26 ships agent_recall_full -- 7-layer federated semantic recall across records, KG, learned rules, CBR episodes, SCM variables, belief tracking, and continual learners in parallel (one question, seven buckets, ~150ms p95). INSTALL: npx -y -p @neruva/mcp@latest neruva-mcp. v0.23 BUNDLES neruva-record-hook native binary (Rust, ~145-340KB) for all 6 platforms. Cache-hit p95 ~80ms via warm daemon + TTL cache. AUTO-PILOT: agent_route_intent_prompt + agent_reflect_prompt + auto-extract on records_ingest. Plus typed Records, federated agent_remember/recall/context, counterfactual rollouts, rule induction, code_kg_* navigation, .neruva V3 container. Pattern-C: substrate stays $0/call. Free tier, no card.",
|
|
5
5
|
"license": "MIT",
|
|
6
6
|
"type": "module",
|