{"id":97149,"date":"2026-09-12T08:00:52","date_gmt":"2026-09-12T05:00:52","guid":{"rendered":"https:\/\/twodots.gr\/?p=97149"},"modified":"2026-09-12T08:00:53","modified_gmt":"2026-09-12T05:00:53","slug":"nexus-mcp-tool-routing-ai-agents","status":"publish","type":"post","link":"https:\/\/twodots.gr\/bg\/nexus-mcp-tool-routing-ai-agents\/","title":{"rendered":"Nexus: \u03c0\u03ce\u03c2 \u03bf\u03b9 AI agents \u03b2\u03c1\u03af\u03c3\u03ba\u03bf\u03c5\u03bd \u03c4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03c0\u03bd\u03af\u03b3\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b1 schemas"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u03a4\u03bf Nexus \u03b5\u03af\u03bd\u03b1\u03b9 \u03ad\u03bd\u03b1 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03b9\u03ba\u03cc prototype \u03c0\u03bf\u03c5 \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf \u03b2\u03ac\u03c1\u03bf\u03c2 \u03c4\u03c9\u03bd tool schemas \u03c3\u03c4\u03bf\u03c5\u03c2 AI agents \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03c2 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5 \u03b1\u03c0\u03cc \u03c4\u03bf \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 schema-prefill.<\/strong> \u03a0\u03c1\u03ce\u03c4\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 semantic retrieval \u03c4\u03c9\u03bd \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b5\u03c1\u03c9\u03bd \u03c5\u03c0\u03bf\u03c8\u03b7\u03c6\u03af\u03c9\u03bd \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 INT8 Semantic Lookaside Buffer \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03ac \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 arguments \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03bc\u03b9\u03b1 \u03c3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b7 \u03ba\u03b5\u03b9\u03bc\u03b5\u03bd\u03b9\u03ba\u03ae \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae. \u03a4\u03bf KV-cache splicing \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2, \u03b1\u03c5\u03c3\u03c4\u03b7\u03c1\u03ac \u03bf\u03c1\u03b9\u03bf\u03b8\u03b5\u03c4\u03b7\u03bc\u03ad\u03bd\u03bf\u03c2 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03c3\u03bc\u03cc\u03c2.<\/p>\n<p>\u03a3\u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf benchmark, \u03c4\u03bf routing \u03ad\u03bc\u03b5\u03b9\u03bd\u03b5 \u03c3\u03c4\u03bf 89% \u03bc\u03b5 registry 250 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd, \u03b5\u03bd\u03ce \u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf argument token \u03ad\u03c6\u03c4\u03b1\u03c3\u03b5 1,66\u00d7 \u03c4\u03b1\u03c7\u03cd\u03c4\u03b5\u03c1\u03b1 \u03b1\u03c0\u03cc full-schema re-prefill \u03bc\u03b5 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 80% \u03bb\u03b9\u03b3\u03cc\u03c4\u03b5\u03c1\u03b1 main-context tokens. \u039f\u03b9 \u03b1\u03c1\u03b9\u03b8\u03bc\u03bf\u03af \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 Qwen2.5-14B configuration \u03c3\u03b5 Apple M4 Max \u03ba\u03b1\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ac \u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1\u00b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03bf envelope, \u03cc\u03c7\u03b9 \u03b3\u03b5\u03bd\u03b9\u03ba\u03ae \u03b5\u03b3\u03b3\u03cd\u03b7\u03c3\u03b7 \u03b3\u03b9\u03b1 cloud APIs \u03ae \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03bf\u03cd\u03c2 \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2.<\/p>\n<\/div>\n<div class=\"td-article-toc\">\n<div class=\"td-toc-title\">\u0421\u044a\u0434\u044a\u0440\u0436\u0430\u043d\u0438\u0435<\/div>\n<ul>\n<li><a href=\"#mcp-registry-prefill-kostos\">\u03a4\u03bf \u03ba\u03c1\u03c5\u03c6\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b5\u03bd\u03cc\u03c2 \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf\u03c5 MCP registry<\/a><\/li>\n<li><a href=\"#retrieval-routing-slb\">Retrieval \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf \u03b2\u03b1\u03c1\u03cd schema<\/a><\/li>\n<li><a href=\"#compressed-tool-signature\">\u03a3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b7 \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae \u03b1\u03bd\u03c4\u03af \u03b3\u03b9\u03b1 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf JSON schema<\/a><\/li>\n<li><a href=\"#nexus-scale-250-tools\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b5 \u03b7 \u03ba\u03bb\u03b9\u03bc\u03ac\u03ba\u03c9\u03c3\u03b7 \u03ad\u03c9\u03c2 250 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1<\/a><\/li>\n<li><a href=\"#kv-cache-rope-drift\">\u03a4\u03bf KV-cache splice \u03ba\u03b1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03b8\u03ad\u03c3\u03b7\u03c2<\/a><\/li>\n<li><a href=\"#depth-adaptive-recompute\">Depth-adaptive recompute \u03ba\u03b1\u03b9 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1<\/a><\/li>\n<li><a href=\"#negative-results\">\u03a4\u03b1 \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ac \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03bf\u03c5 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b1\u03bd \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae<\/a><\/li>\n<li><a href=\"#hardware-boundaries\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf hardware boundary \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c5\u03c0\u03bf\u03c3\u03b7\u03bc\u03b5\u03af\u03c9\u03c3\u03b7<\/a><\/li>\n<li><a href=\"#nexus-ai-workflows\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03b6\u03bf\u03c5\u03bd AI workflows<\/a><\/li>\n<li><a href=\"#nexus-evaluation-pilot\">\u03a0\u03ce\u03c2 \u03bd\u03b1 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03c3\u03b5\u03c4\u03b5 \u03bc\u03b9\u03b1 \u03c0\u03b1\u03c1\u03cc\u03bc\u03bf\u03b9\u03b1 \u03bb\u03cd\u03c3\u03b7<\/a><\/li>\n<li><a href=\"#nexus-main-lesson\">\u0397 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1 \u03c4\u03bf\u03c5 Nexus<\/a><\/li>\n<\/ul>\n<\/div>\n<h2 id=\"mcp-registry-prefill-kostos\">\u03a4\u03bf \u03ba\u03c1\u03c5\u03c6\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b5\u03bd\u03cc\u03c2 \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf\u03c5 MCP registry<\/h2>\n<p>\u0388\u03bd\u03b1\u03c2 AI agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9 \u03b4\u03b5\u03ba\u03ac\u03b4\u03b5\u03c2 \u03ae \u03b5\u03ba\u03b1\u03c4\u03bf\u03bd\u03c4\u03ac\u03b4\u03b5\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u2019 \u03cc\u03bb\u03b1 \u03b1\u03c5\u03c4\u03ac \u03bd\u03b1 \u03ba\u03b1\u03b8\u03c5\u03c3\u03c4\u03b5\u03c1\u03b5\u03af \u03c0\u03c1\u03b9\u03bd \u03ba\u03ac\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03c0\u03c1\u03ce\u03c4\u03b7 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03ba\u03af\u03bd\u03b7\u03c3\u03b7. \u0397 \u03b1\u03b9\u03c4\u03af\u03b1 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03b7 \u00ab\u03c3\u03ba\u03ad\u03c8\u03b7\u00bb \u03c4\u03bf\u03c5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5. \u03a3\u03c5\u03c7\u03bd\u03ac \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03b2\u03ac\u03c1\u03bf\u03c2 \u03c4\u03c9\u03bd tool schemas \u03c0\u03bf\u03c5 \u03b5\u03c0\u03b1\u03bd\u03b5\u03b9\u03c3\u03ac\u03b3\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf prompt \u03ba\u03b1\u03b9 \u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03be\u03b1\u03bd\u03ac \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03b3\u03cd\u03c1\u03bf.<\/p>\n<p>\u03a3\u03c4\u03bf Model Context Protocol \u03ba\u03ac\u03b8\u03b5 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03cc\u03bd\u03bf\u03bc\u03b1, \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ae \u03ba\u03b1\u03b9 JSON Schema \u03b3\u03b9\u03b1 \u03c4\u03b1 arguments \u03c4\u03bf\u03c5. \u038c\u03c3\u03bf \u03b1\u03c5\u03be\u03ac\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1, \u03b1\u03c5\u03c4\u03ad\u03c2 \u03bf\u03b9 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2 \u03ba\u03b1\u03c4\u03b1\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf \u03c4\u03bc\u03ae\u03bc\u03b1 \u03c4\u03bf\u03c5 prompt. \u03a4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c4\u03b9\u03c2 \u03c0\u03b5\u03c1\u03ac\u03c3\u03b5\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf prefill \u03c0\u03c1\u03b9\u03bd \u03c0\u03b1\u03c1\u03b1\u03b3\u03ac\u03b3\u03b5\u03b9 \u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf token \u03c4\u03b7\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2 \u03ae \u03c4\u03b7\u03c2 \u03ba\u03bb\u03ae\u03c3\u03b7\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5.<\/p>\n<p>\u03a4\u03bf self-attention \u03ad\u03c7\u03b5\u03b9 \u03c4\u03b5\u03c4\u03c1\u03b1\u03b3\u03c9\u03bd\u03b9\u03ba\u03ae \u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c3\u03b7 \u03b1\u03c0\u03cc \u03c4\u03bf \u03bc\u03ae\u03ba\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03af\u03b1\u03c2. \u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 <em>Nexus: Depth-Adaptive KV-Cache Splicing and Retrieval-Decoupled Tool Routing for Agentic LLMs on Unified Memory<\/em> \u03b4\u03b5\u03bd \u03c0\u03b1\u03c1\u03bf\u03c5\u03c3\u03b9\u03ac\u03b6\u03b5\u03b9 \u03b1\u03c5\u03c4\u03ae \u03c4\u03b7 \u03c3\u03c7\u03ad\u03c3\u03b7 \u03c9\u03c2 \u03bd\u03ad\u03bf \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1\u00b7 \u03c4\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03b5\u03be\u03b7\u03b3\u03ae\u03c3\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03b7 \u03b5\u03c0\u03b1\u03bd\u03b1\u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03c0\u03bf\u03bb\u03bb\u03ce\u03bd verbose schemas \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03ad\u03bd\u03b1 \u00abprefill wall\u00bb. \u03a4\u03bf \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b5\u03bd\u03c4\u03bf\u03bd\u03cc\u03c4\u03b5\u03c1\u03bf \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf \u03b9\u03c3\u03c4\u03bf\u03c1\u03b9\u03ba\u03cc \u03bc\u03b5\u03b3\u03b1\u03bb\u03ce\u03bd\u03b5\u03b9 \u03ba\u03b1\u03b9 \u03c4\u03bf registry \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03b5\u03ba\u03b1\u03c4\u03bf\u03bd\u03c4\u03ac\u03b4\u03b5\u03c2 \u03c0\u03b9\u03b8\u03b1\u03bd\u03ad\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2.<\/p>\n<p>\u039c\u03b9\u03b1 \u03b1\u03c0\u03bb\u03ae \u03c0\u03c1\u03bf\u03c3\u03ad\u03b3\u03b3\u03b9\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03c4\u03bf\u03c0\u03bf\u03b8\u03b5\u03c4\u03b7\u03b8\u03bf\u03cd\u03bd \u03cc\u03bb\u03b1 \u03c4\u03b1 schemas \u03c3\u03c4\u03bf context. \u039c\u03b9\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03c4\u03b7\u03b8\u03bf\u03cd\u03bd \u03c0\u03c1\u03ce\u03c4\u03b1 \u03bb\u03af\u03b3\u03b1 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b3\u03af\u03bd\u03b5\u03b9 prefill \u03bc\u03cc\u03bd\u03bf \u03c3\u03b5 \u03b1\u03c5\u03c4\u03ac. \u03a4\u03bf Nexus \u03c0\u03b7\u03b3\u03b1\u03af\u03bd\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03b2\u03ae\u03bc\u03b1 \u03c0\u03b9\u03bf \u03c0\u03ad\u03c1\u03b1: \u03b1\u03c0\u03bf\u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03b9 \u03c4\u03bf semantic routing \u03b1\u03c0\u03cc \u03c4\u03bf schema-prefill \u03ba\u03b1\u03b9 \u03ba\u03c1\u03b1\u03c4\u03ac \u03c4\u03b7\u03bd \u03c0\u03bb\u03ae\u03c1\u03b7 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ae \u03ad\u03be\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03ba\u03cd\u03c1\u03b9\u03bf \u03bc\u03bf\u03bd\u03bf\u03c0\u03ac\u03c4\u03b9 \u03cc\u03c0\u03bf\u03c5 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af. \u0397 \u03af\u03b4\u03b9\u03b1 \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b3\u03b9\u03b1 \u03c3\u03b1\u03c6\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03ba\u03b1\u03b9 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03b5\u03be\u03b5\u03c4\u03ac\u03b6\u03bf\u03c5\u03bc\u03b5 <a href=\"https:\/\/twodots.gr\/ai-agents-poios-ftaiei-otan-aftomatopoiisi-apotygchanei\/\">\u03c0\u03bf\u03b9\u03bf\u03c2 \u03b5\u03c5\u03b8\u03cd\u03bd\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03ad\u03bd\u03b1\u03c2 AI agent \u03b1\u03c0\u03bf\u03c4\u03c5\u03b3\u03c7\u03ac\u03bd\u03b5\u03b9<\/a>.<\/p>\n<div class=\"td-comparison td-comparison-cards td-comparison-cards--horizontal\">\n<div class=\"td-comparison-grid td-comparison-grid--three\">\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">\u038c\u03bb\u03b1 \u03c4\u03b1 schemas \u03c3\u03c4\u03bf prompt<\/p>\n<p>\u0397 \u03b1\u03c0\u03bb\u03bf\u03cd\u03c3\u03c4\u03b5\u03c1\u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae, \u03b1\u03bb\u03bb\u03ac \u03c4\u03bf prefill \u03bc\u03b5\u03b3\u03b1\u03bb\u03ce\u03bd\u03b5\u03b9 \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 \u03c4\u03bf registry \u03ba\u03b1\u03b9 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03be\u03b1\u03bd\u03c4\u03bb\u03ae\u03c3\u03b5\u03b9 \u03c4\u03bf context window.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Full registry<\/span><span class=\"td-badge\">\u03a5\u03c8\u03b7\u03bb\u03cc prefill<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">Retrieve-K \u03ba\u03b1\u03b9 prefill<\/p>\n<p>\u0391\u03bd\u03b1\u03ba\u03c4\u03ac \u03bb\u03af\u03b3\u03bf\u03c5\u03c2 \u03c5\u03c0\u03bf\u03c8\u03b7\u03c6\u03af\u03bf\u03c5\u03c2, \u03b1\u03bb\u03bb\u03ac \u03b5\u03be\u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03bd\u03b1 \u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03b5\u03af online \u03c4\u03b1 \u03c0\u03bb\u03ae\u03c1\u03b7 schemas \u03c0\u03bf\u03c5 \u03b5\u03c0\u03ad\u03bb\u03b5\u03be\u03b5.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Top-K<\/span><span class=\"td-badge\">\u039c\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf context<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card td-platform-card--navy\">\n<p class=\"td-comparison-title\">Nexus hybrid path<\/p>\n<p>\u039a\u03ac\u03bd\u03b5\u03b9 routing \u03ad\u03be\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b2\u03b1\u03b8\u03cd context \u03ba\u03b1\u03b9 \u03b4\u03af\u03bd\u03b5\u03b9 \u03c3\u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03b9\u03b1 \u03bc\u03b9\u03ba\u03c1\u03ae type-hinted \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae \u03b3\u03b9\u03b1 \u03c4\u03b1 arguments.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">INT8 SLB<\/span><span class=\"td-badge\">Compressed IR<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<h2 id=\"retrieval-routing-slb\">Retrieval \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf \u03b2\u03b1\u03c1\u03cd schema<\/h2>\n<p>\u03a4\u03bf Nexus \u03b1\u03c0\u03bf\u03b8\u03b7\u03ba\u03b5\u03cd\u03b5\u03b9 embeddings \u03b3\u03b9\u03b1 \u03c4\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03c3\u03b5 \u03ad\u03bd\u03b1 INT8 Semantic Lookaside Buffer. \u0393\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03b6\u03b5\u03b9 embedding \u03c4\u03bf\u03c5 query \u03ba\u03b1\u03b9 \u03c3\u03b1\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf registry \u03bc\u03b5 SIMD dot products. \u03a4\u03bf \u03cc\u03bd\u03bf\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03b7 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ae \u03ba\u03ac\u03b8\u03b5 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5 \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b7\u03bd \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7, \u03bc\u03b5 \u03c4\u03bf \u03cc\u03bd\u03bf\u03bc\u03b1 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9, \u03c3\u03cd\u03bc\u03c6\u03c9\u03bd\u03b1 \u03bc\u03b5 \u03c4\u03bf paper, \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc routing signal.<\/p>\n<p>\u039f\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03b9 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03bf\u03b9 \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc margin gate. \u038c\u03c4\u03b1\u03bd \u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03c4\u03b1 \u03b4\u03cd\u03bf \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b1 scores \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c1\u03ba\u03b5\u03c4\u03ae, \u03b7 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b1. \u038c\u03c4\u03b1\u03bd \u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ae, \u03b5\u03bd\u03b5\u03c1\u03b3\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1\u03c2 fine-tuned cross-encoder \u03c4\u03cd\u03c0\u03bf\u03c5 MiniLM \u03b3\u03b9\u03b1 \u03c4\u03b7 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b7 \u03c0\u03b5\u03c1\u03af\u03c0\u03c4\u03c9\u03c3\u03b7. \u03a4\u03bf threshold \u03bf\u03c1\u03af\u03c3\u03c4\u03b7\u03ba\u03b5 \u03c3\u03c4\u03bf 20\u03cc \u03b5\u03ba\u03b1\u03c4\u03bf\u03c3\u03c4\u03b7\u03bc\u03cc\u03c1\u03b9\u03bf adversarial margins, \u03c3\u03c4\u03bf 0,0136, \u03ba\u03b1\u03b9 \u03c3\u03c4\u03bf benchmark \u03bf cross-encoder \u03b5\u03bd\u03b5\u03c1\u03b3\u03bf\u03c0\u03bf\u03b9\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c3\u03c4\u03bf 20,8% \u03c4\u03c9\u03bd \u03b1\u03c0\u03bf\u03c6\u03ac\u03c3\u03b5\u03c9\u03bd.<\/p>\n<p>\u03a4\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf \u03c7\u03b1\u03c1\u03b1\u03ba\u03c4\u03b7\u03c1\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf routing \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf query \u03ba\u03b1\u03b9 \u03c4\u03bf registry, \u03cc\u03c7\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b2\u03ac\u03b8\u03bf\u03c2 \u03c4\u03bf\u03c5 live context. \u0386\u03c1\u03b1 \u03b4\u03b5\u03bd \u03b5\u03c0\u03b9\u03b2\u03b1\u03c1\u03cd\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03af\u03b4\u03b9\u03bf positional \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03b7 \u03bc\u03b5\u03c4\u03b1\u03c6\u03bf\u03c1\u03ac KV cache. \u0391\u03c5\u03c4\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03c0\u03b9\u03bf \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03c3\u03b9\u03bc\u03b7 \u03b1\u03c1\u03c7\u03ae \u03c4\u03bf\u03c5 Nexus \u03b1\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03b3\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03b4\u03bf\u03c5\u03bb\u03b5\u03cd\u03bf\u03c5\u03bd \u03bc\u03b5 cloud \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03ae \u03bc\u03b5 <a href=\"https:\/\/twodots.gr\/no-code-ai-ergaleia-rag-agents-workflows\/\">no-code \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03b3\u03b9\u03b1 RAG, agents \u03ba\u03b1\u03b9 workflows<\/a>.<\/p>\n<h2 id=\"compressed-tool-signature\">\u03a3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b7 \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae \u03b1\u03bd\u03c4\u03af \u03b3\u03b9\u03b1 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf JSON schema<\/h2>\n<p>\u0391\u03c6\u03bf\u03cd \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03b5\u03af \u03c4\u03bf \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf, \u03bf agent \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03ae\u03c3\u03b5\u03b9 \u03c4\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ac arguments. \u03a4\u03bf Nexus \u03b4\u03b5\u03bd \u03b5\u03c0\u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c5\u03c0\u03bf\u03c7\u03c1\u03b5\u03c9\u03c4\u03b9\u03ba\u03ac \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf \u03c4\u03bf schema \u03c3\u03c4\u03bf \u03ba\u03cd\u03c1\u03b9\u03bf context. \u03a4\u03bf \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c3\u03b5 \u03c3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b7, type-hinted \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae, \u03b3\u03b9\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1 \u03bc\u03b9\u03b1 \u03bc\u03bf\u03c1\u03c6\u03ae \u03cc\u03c0\u03c9\u03c2 <code>create_repository(name: string, private?: boolean)<\/code>, \u03bc\u03b5 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2 \u03cc\u03c0\u03bf\u03c5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9.<\/p>\n<p>\u03a3\u03c4\u03bf \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03bf \u03c3\u03cd\u03bd\u03bf\u03bb\u03bf \u03b7 \u03b4\u03b9\u03ac\u03bc\u03b5\u03c3\u03b7 \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae \u03b5\u03af\u03c7\u03b5 19 tokens \u03ba\u03b1\u03b9 \u03c4\u03bf 99\u03bf \u03b5\u03ba\u03b1\u03c4\u03bf\u03c3\u03c4\u03b7\u03bc\u03cc\u03c1\u03b9\u03bf 32. \u0397 \u03c3\u03c5\u03bc\u03c0\u03af\u03b5\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf prefill \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af \u03c4\u03b1 \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5\u03c2 \u03c4\u03cd\u03c0\u03bf\u03c5\u03c2 \u03c4\u03c9\u03bd \u03c0\u03b5\u03b4\u03af\u03c9\u03bd \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf. \u0388\u03bd\u03b1 finite-state machine \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03bf\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5, \u03b5\u03bd\u03ce grammar \u03b1\u03c0\u03cc \u03c4\u03bf schema \u03b5\u03c0\u03b9\u03b2\u03ac\u03bb\u03bb\u03b5\u03b9 \u03ad\u03b3\u03ba\u03c5\u03c1\u03b7 JSON \u03b4\u03bf\u03bc\u03ae \u03c3\u03c4\u03b1 arguments.<\/p>\n<p>\u03a3\u03c4\u03b9\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03cc\u03c0\u03bf\u03c5 \u03c4\u03bf routing \u03ae\u03c4\u03b1\u03bd \u03c3\u03c9\u03c3\u03c4\u03cc, \u03c3\u03c5\u03bc\u03c0\u03bb\u03b7\u03c1\u03ce\u03b8\u03b7\u03ba\u03b5 \u03c3\u03c9\u03c3\u03c4\u03ac \u03c4\u03bf 100% \u03c4\u03c9\u03bd 40 query-specified arguments, \u03bc\u03b5 \u03ba\u03ac\u03c4\u03c9 \u03cc\u03c1\u03b9\u03bf 95% confidence interval 91,2%. \u0397 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03ae end-to-end \u03b1\u03ba\u03c1\u03af\u03b2\u03b5\u03b9\u03b1 arguments, \u03cc\u03c4\u03b1\u03bd \u03c4\u03b1 routing errors \u03bc\u03b5\u03c4\u03c1\u03bf\u03cd\u03bd \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03ae\u03c4\u03b1\u03bd 80% \u03c3\u03b5 50 arguments. \u0397 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae: \u03ad\u03b3\u03ba\u03c5\u03c1\u03bf JSON \u03b4\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1 \u03b1\u03bd \u03ad\u03c7\u03b5\u03b9 \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03b5\u03af \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf.<\/p>\n<aside class=\"td-article-note\">\n<p><strong>\u039c\u03b7\u03bd \u03c3\u03c5\u03b3\u03c7\u03ad\u03b5\u03c4\u03b5 \u03c4\u03b7 \u03c3\u03cd\u03bd\u03c4\u03b1\u03be\u03b7 \u03bc\u03b5 \u03c4\u03b7\u03bd \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1:<\/strong> \u03c4\u03bf Nexus \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 100% JSON validity \u03c3\u03c4\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc consistency set, \u03b1\u03bb\u03bb\u03ac 80% end-to-end argument accuracy \u03cc\u03c4\u03b1\u03bd \u03c7\u03c1\u03b5\u03ce\u03b8\u03b7\u03ba\u03b1\u03bd \u03ba\u03b1\u03b9 \u03c4\u03b1 routing errors. \u03a3\u03b5 CRM, ERP \u03ae publishing workflow, \u03c4\u03bf \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03bc\u03b5 \u03ac\u03c8\u03bf\u03b3\u03bf payload \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1.<\/p>\n<\/aside>\n<div class=\"td-chart td-chart--metrics\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-title\">\u03a4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 \u03c4\u03bf\u03c5 Nexus<\/p>\n<p class=\"td-chart-subtitle\">\u0391\u03c6\u03bf\u03c1\u03bf\u03cd\u03bd \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf GitHub-MCP benchmark \u03ba\u03b1\u03b9 \u03c4\u03bf \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03bf M4 Max\/Qwen2.5-14B configuration, \u03cc\u03c7\u03b9 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03ac KPIs.<\/p>\n<\/div>\n<div class=\"td-chart-body\">\n<div class=\"td-metric-grid\">\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">250<\/span><span class=\"td-metric-label\">\u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03c3\u03c4\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf registry<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">89%<\/span><span class=\"td-metric-label\">end-to-end routing accuracy<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">19<\/span><span class=\"td-metric-label\">tokens \u03c3\u03c4\u03b7 \u03b4\u03b9\u03ac\u03bc\u03b5\u03c3\u03b7 \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">1,66\u00d7<\/span><span class=\"td-metric-label\">\u03c4\u03b1\u03c7\u03cd\u03c4\u03b5\u03c1\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf argument token<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<h2 id=\"nexus-scale-250-tools\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b5 \u03b7 \u03ba\u03bb\u03b9\u03bc\u03ac\u03ba\u03c9\u03c3\u03b7 \u03ad\u03c9\u03c2 250 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1<\/h2>\n<p>\u03a4\u03bf routing \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c3\u03b5 100 GitHub-MCP queries \u03bc\u03b5 registries 10, 50, 100 \u03ba\u03b1\u03b9 250 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd. \u0397 end-to-end accuracy \u03ae\u03c4\u03b1\u03bd \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b1 92%, 90%, 89% \u03ba\u03b1\u03b9 89%. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03bf\u03c5\u03bd \u03c4\u03b7\u03bd \u03b5\u03c0\u03af\u03b4\u03bf\u03c3\u03b7 \u03c9\u03c2 \u00ab\u03c3\u03c7\u03b5\u03b4\u03cc\u03bd \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03b7\u00bb, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c4\u03b1 confidence intervals \u03b5\u03c0\u03b9\u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03bf\u03bd\u03c4\u03b1\u03b9, \u03cc\u03c7\u03b9 \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03b4\u03b5\u03bd \u03ad\u03c7\u03b5\u03b9 \u03ba\u03b1\u03bc\u03af\u03b1 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7.<\/p>\n<p>\u03a3\u03c4\u03b1 250 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1, \u03c4\u03bf 89% \u03b5\u03af\u03c7\u03b5 Wilson 95% confidence interval \u03b1\u03c0\u03cc 81,4% \u03ad\u03c9\u03c2 93,7%. \u03a4\u03bf SLB \u03b5\u03af\u03c7\u03b5 top-1 recall 74% \u03ba\u03b1\u03b9 top-3 recall 95%. \u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03bf \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03c3\u03bc\u03cc\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf\u03c2: \u03b7 \u03b3\u03c1\u03ae\u03b3\u03bf\u03c1\u03b7 semantic \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7 \u03c3\u03c5\u03c7\u03bd\u03ac \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03b9 \u03c4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf\u03c5\u03c2 \u03ba\u03bf\u03c1\u03c5\u03c6\u03b1\u03af\u03bf\u03c5\u03c2 \u03c5\u03c0\u03bf\u03c8\u03b7\u03c6\u03af\u03bf\u03c5\u03c2, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c7\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03c3\u03c4\u03b7\u03bd \u03c0\u03c1\u03ce\u03c4\u03b7 \u03b8\u03ad\u03c3\u03b7.<\/p>\n<p>\u039f baseline \u03c0\u03bf\u03c5 \u03c3\u03c5\u03bd\u03ad\u03b4\u03b5\u03b5 \u03cc\u03bb\u03b1 \u03c4\u03b1 schemas \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 98% \u03c3\u03c4\u03b1 10 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1, \u03cc\u03bc\u03c9\u03c2 \u03b1\u03c0\u03cc \u03c4\u03b1 50 \u03ba\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c9 \u03be\u03b5\u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03c3\u03b5 \u03c4\u03bf \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03bf context \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03c3\u03b5 \u03bd\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 routing. \u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03ae \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03bc\u03b5 \u03ba\u03ac\u03b8\u03b5 agent framework. \u0395\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7\u03c2 \u03c0\u03b5\u03b9\u03c1\u03b1\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae\u03c2 \u03c1\u03cd\u03b8\u03bc\u03b9\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b1\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03c4\u03bf scaling failure \u03c4\u03bf\u03c5 \u00ab\u03b2\u03ac\u03bb\u03b5 \u03c4\u03b1 \u03cc\u03bb\u03b1 \u03c3\u03c4\u03bf prompt\u00bb.<\/p>\n<p>\u0397 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b1\u03bd\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u00ab\u03b1\u03b3\u03bf\u03c1\u03ac\u03c3\u03c4\u03b5 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u00bb, \u03b1\u03bb\u03bb\u03ac \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf registry: \u03b4\u03b9\u03b1\u03ba\u03c1\u03b9\u03c4\u03ac \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03b1, \u03c3\u03b1\u03c6\u03b5\u03af\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2, \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 permissions \u03ba\u03b1\u03b9 tests \u03b3\u03b9\u03b1 \u03c3\u03c5\u03b3\u03b3\u03b5\u03bd\u03b9\u03ba\u03ac \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1. \u0391\u03c5\u03c4\u03ae \u03b7 \u03c0\u03b5\u03b9\u03b8\u03b1\u03c1\u03c7\u03af\u03b1 \u03c3\u03c5\u03bc\u03c0\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03ad\u03bd\u03b1 <a href=\"https:\/\/twodots.gr\/full-stack-ai-integrated-digital-products\/\">full-stack AI \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd<\/a>, \u03cc\u03c0\u03bf\u03c5 orchestration, \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1, monitoring \u03ba\u03b1\u03b9 \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03af\u03b1 \u03c7\u03c1\u03ae\u03c3\u03c4\u03b7 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03bf\u03cd\u03bd \u03c9\u03c2 \u03ad\u03bd\u03b1 \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1.<\/p>\n<h2 id=\"kv-cache-rope-drift\">\u03a4\u03bf KV-cache splice \u03ba\u03b1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03b8\u03ad\u03c3\u03b7\u03c2<\/h2>\n<p>\u0397 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b9\u03b4\u03ad\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03b3\u03bb\u03c9\u03c4\u03c4\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 schema \u03bc\u03af\u03b1 \u03c6\u03bf\u03c1\u03ac \u03c3\u03b5 block \u03b1\u03c0\u03cc keys \u03ba\u03b1\u03b9 values \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03b9\u03c3\u03ac\u03b3\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03b5\u03c5\u03b8\u03b5\u03af\u03b1\u03c2 \u03c3\u03c4\u03bf live KV cache. \u0391\u03bd \u03b1\u03c5\u03c4\u03cc \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03bf\u03cd\u03c3\u03b5 \u03c3\u03b5 \u03bf\u03c0\u03bf\u03b9\u03b1\u03b4\u03ae\u03c0\u03bf\u03c4\u03b5 \u03b8\u03ad\u03c3\u03b7, \u03bf agent \u03b8\u03b1 \u03b1\u03c0\u03ad\u03c6\u03b5\u03c5\u03b3\u03b5 \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03bf\u03c5 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03b1\u03bd\u03cc\u03bc\u03b5\u03bd\u03bf\u03c5 prefill. \u0397 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03af\u03b1 \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf Rotary Position Embedding: \u03c4\u03b1 keys \u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd \u03c4\u03b7 \u03b8\u03ad\u03c3\u03b7 \u03c3\u03c4\u03b7\u03bd \u03bf\u03c0\u03bf\u03af\u03b1 \u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03c3\u03c4\u03b7\u03ba\u03b1\u03bd.<\/p>\n<p>\u0388\u03bd\u03b1 block \u03c0\u03bf\u03c5 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c3\u03b5 anchor position \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03b7\u03b8\u03b5\u03af \u03b1\u03c5\u03b8\u03b1\u03af\u03c1\u03b5\u03c4\u03b1 \u03c3\u03b5 \u03b2\u03b1\u03b8\u03cd\u03c4\u03b5\u03c1\u03bf context \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03c4\u03b7\u03c2 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae\u03c2 attention. \u03a4\u03bf Nexus \u03b5\u03c0\u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c4\u03b7 RoPE \u03c6\u03ac\u03c3\u03b7 \u03c3\u03c4\u03b7 \u03bd\u03ad\u03b1 \u03b8\u03ad\u03c3\u03b7, \u03cc\u03bc\u03c9\u03c2 \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ae \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7 \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c4\u03bf block \u03b5\u03af\u03c7\u03b5 \u03c0\u03b1\u03c1\u03b1\u03c7\u03b8\u03b5\u03af \u03ba\u03ac\u03c4\u03c9 \u03b1\u03c0\u03cc \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf context. \u03a3\u03c4\u03bf anchor \u03c4\u03bf output \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ad\u03c2. \u0395\u03ba\u03c4\u03cc\u03c2 anchor, \u03b7 measured divergence \u03ba\u03b9\u03bd\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c3\u03c4\u03b7 \u03b6\u03ce\u03bd\u03b7 0,008\u20130,038 nats \u03bc\u03ad\u03c7\u03c1\u03b9 offset 2.048 tokens, \u03bc\u03b5 top-1 agreement 1,0.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03bd \u03b8\u03b5\u03c9\u03c1\u03ae\u03b8\u03b7\u03ba\u03b5 \u03b1\u03c1\u03ba\u03b5\u03c4\u03cc \u03b3\u03b9\u03b1 \u03c3\u03c4\u03bf\u03c7\u03b1\u03c3\u03c4\u03b9\u03ba\u03cc decoding \u03c0\u03bf\u03bb\u03bb\u03ce\u03bd \u03b2\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd. \u0391\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ae \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7 \u03c3\u03c4\u03b7\u03bd \u03bf\u03c5\u03c1\u03ac \u03c4\u03b7\u03c2 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c3\u03c5\u03c3\u03c3\u03c9\u03c1\u03b5\u03c5\u03c4\u03b5\u03af \u03c3\u03b5 \u03bc\u03b9\u03b1 \u03c0\u03bf\u03bb\u03c5\u03b2\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae agentic \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae. \u0393\u03b9\u2019 \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf Nexus \u03b4\u03b5\u03bd \u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 \u00ab\u03b3\u03c5\u03bc\u03bd\u03cc\u00bb splice \u03c0\u03ad\u03c1\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf \u03cc\u03c1\u03b9\u03bf \u03c4\u03c9\u03bd 256 tokens, \u03c0\u03b1\u03c1\u03cc\u03c4\u03b9 \u03c4\u03bf \u03cc\u03c1\u03b9\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03bd\u03c4\u03b7\u03c1\u03b7\u03c4\u03b9\u03ba\u03ae \u03ad\u03bd\u03b1\u03c1\u03be\u03b7 \u03b5\u03c0\u03b9\u03b4\u03b9\u03cc\u03c1\u03b8\u03c9\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03b1\u03c0\u03cc\u03c4\u03bf\u03bc\u03bf\u03c2 \u03b3\u03ba\u03c1\u03b5\u03bc\u03cc\u03c2 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2.<\/p>\n<h2 id=\"depth-adaptive-recompute\">Depth-adaptive recompute \u03ba\u03b1\u03b9 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1<\/h2>\n<p>\u03a0\u03ac\u03bd\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03cc\u03c1\u03b9\u03bf, \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03b1\u03ba\u03c5\u03c1\u03ce\u03bd\u03b5\u03b9 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03ad\u03bd\u03b1 \u03c3\u03c5\u03bd\u03b5\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf suffix \u03c4\u03bf\u03c5 schema. \u03a4\u03bf \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03b1\u03c0\u03cc 5% \u03ba\u03b1\u03b9 \u03b1\u03c5\u03be\u03ac\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03bf \u03b2\u03ac\u03b8\u03bf\u03c2 \u03bc\u03ad\u03c7\u03c1\u03b9 \u03bd\u03b1 \u03c6\u03c4\u03ac\u03c3\u03b5\u03b9 \u03c3\u03b5 \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 re-prefill. \u039c\u03b5 \u03b1\u03c5\u03c4\u03cc\u03bd \u03c4\u03bf\u03bd \u03c4\u03c1\u03cc\u03c0\u03bf \u03b7 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae \u03c4\u03bf\u03c5 \u03b5\u03c0\u03cc\u03bc\u03b5\u03bd\u03bf\u03c5 token \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03b5\u03b9 \u03c3\u03c4\u03b7 reference \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac, \u03bc\u03b5 top-1 agreement \u03ba\u03b1\u03b9 KL divergence \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 \u03bc\u03b7\u03b4\u03ad\u03bd.<\/p>\n<p>\u0397 \u03b9\u03b4\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u00abnever-regress\u00bb \u03b1\u03c6\u03bf\u03c1\u03ac \u03c4\u03b7\u03bd \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 output, \u03cc\u03c7\u03b9 \u03b5\u03b3\u03b3\u03c5\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c0\u03b9\u03c4\u03ac\u03c7\u03c5\u03bd\u03c3\u03b7. \u03a3\u03c4\u03b7\u03bd default \u03ba\u03b1\u03bc\u03c0\u03cd\u03bb\u03b7, \u03c3\u03c4\u03b1 256 tokens \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c6\u03b7\u03ba\u03b5 speedup 1,63\u00d7 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b1 512 tokens 1,24\u00d7. \u03a3\u03c4\u03b1 1.024 tokens \u03b7 \u03b1\u03bd\u03b1\u03bb\u03bf\u03b3\u03af\u03b1 \u03ad\u03c0\u03b5\u03c3\u03b5 \u03c3\u03b5 0,98\u00d7 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b1 2.048 \u03c3\u03c5\u03bd\u03ad\u03ba\u03bb\u03b9\u03bd\u03b5 \u03c3\u03b5 1,00\u00d7. \u039c\u03b9\u03b1 \u03c0\u03b9\u03bf \u03ae\u03c0\u03b9\u03b1 tuned \u03ba\u03b1\u03bc\u03c0\u03cd\u03bb\u03b7 \u03b4\u03b9\u03b1\u03c4\u03ae\u03c1\u03b7\u03c3\u03b5 speedup 1,73\u00d7, 1,42\u00d7, 1,17\u00d7 \u03ba\u03b1\u03b9 1,07\u00d7 \u03c3\u03c4\u03b1 \u03af\u03b4\u03b9\u03b1 \u03b2\u03ac\u03b8\u03b7, \u03b1\u03bb\u03bb\u03ac \u03bc\u03b5\u03c4\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c4\u03bf \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 recompute \u03b1\u03c1\u03b3\u03cc\u03c4\u03b5\u03c1\u03b1.<\/p>\n<p>\u03a4\u03bf trade-off \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03b8\u03b1\u03c1\u03cc: \u03cc\u03c3\u03bf \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03b2\u03b1\u03b8\u03cd context, \u03c4\u03cc\u03c3\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b7 \u03b5\u03be\u03bf\u03b9\u03ba\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7. \u03a4\u03bf paper \u03b4\u03b5\u03bd \u03ba\u03c1\u03cd\u03b2\u03b5\u03b9 \u03b1\u03c5\u03c4\u03ae \u03c4\u03b7\u03bd \u03c5\u03c0\u03bf\u03c7\u03ce\u03c1\u03b7\u03c3\u03b7. \u03a4\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03bf\u03c1\u03b9\u03bf\u03b8\u03b5\u03c4\u03ae\u03c3\u03b5\u03b9 \u03c0\u03bf\u03cd \u03c4\u03bf splicing \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03c0\u03bb\u03b5\u03bf\u03bd\u03ad\u03ba\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03bf\u03cd \u03c4\u03bf \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 fallback \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c7\u03b5\u03b4\u03cc\u03bd \u03b9\u03c3\u03bf\u03b4\u03cd\u03bd\u03b1\u03bc\u03bf \u03bc\u03b5 \u03ba\u03b1\u03bd\u03bf\u03bd\u03b9\u03ba\u03cc prefill.<\/p>\n<h2 id=\"negative-results\">\u03a4\u03b1 \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ac \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03bf\u03c5 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b1\u03bd \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae<\/h2>\n<p>\u039c\u03af\u03b1 \u03c0\u03b1\u03bb\u03b1\u03b9\u03cc\u03c4\u03b5\u03c1\u03b7 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ae scattered partial recompute, \u03bc\u03b5 \u03c4\u03b7\u03bd \u03bf\u03bd\u03bf\u03bc\u03b1\u03c3\u03af\u03b1 LegoLink, \u03ac\u03c6\u03b7\u03c3\u03b5 divergence \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 5,7 nats \u03c3\u03b5 \u03b2\u03ac\u03b8\u03bf\u03c2 1.024, \u03c0\u03bf\u03bb\u03cd \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03cc \u03c4\u03bf \u03c3\u03c5\u03bd\u03b5\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf suffix repair. \u03a4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b5 \u03c3\u03c4\u03b7\u03bd \u03b1\u03c0\u03cc\u03c3\u03c5\u03c1\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03b4\u03b9\u03ac\u03c3\u03c0\u03b1\u03c1\u03c4\u03b7\u03c2 \u03b5\u03c0\u03b9\u03b4\u03b9\u03cc\u03c1\u03b8\u03c9\u03c3\u03b7\u03c2. \u0395\u03af\u03bd\u03b1\u03b9 \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03cc \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1 \u03c4\u03bf\u03c5 \u03b3\u03b9\u03b1\u03c4\u03af \u03ad\u03bd\u03b1 \u00ab\u03ad\u03be\u03c5\u03c0\u03bd\u03bf\u00bb shortcut \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae \u03c7\u03b5\u03b9\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b1\u03bd\u03c4\u03af \u03bd\u03b1 \u03c4\u03b7 \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03c3\u03b5\u03b9.<\/p>\n<p>\u0391\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03ad\u03bd\u03b1\u03c2 \u03c6\u03b8\u03b7\u03bd\u03cc\u03c2 reference-free gate \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03c3\u03b5 K-variance \u03c4\u03bf\u03c5 \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf\u03c5 context \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03c0\u03c1\u03bf\u03b2\u03bb\u03ad\u03c8\u03b5\u03b9 \u03c0\u03bf\u03b9\u03b1 attention heads \u03b8\u03b1 \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03c3\u03bf\u03c5\u03bd drift. \u0397 \u03bc\u03ad\u03c3\u03b7 Spearman correlation \u03ae\u03c4\u03b1\u03bd 0,193, \u03ba\u03ac\u03c4\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03c3\u03c4\u03cc\u03c7\u03bf 0,40. \u03a4\u03bf drift \u03c5\u03c0\u03ae\u03c1\u03c7\u03b5, \u03b1\u03bb\u03bb\u03ac \u03bf proxy \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03c3\u03b5 \u03bd\u03b1 \u03ba\u03b1\u03c4\u03b1\u03c4\u03ac\u03be\u03b5\u03b9 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b1 \u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2.<\/p>\n<p>\u0391\u03c5\u03c4\u03ac \u03c4\u03b1 \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ac \u03b5\u03c5\u03c1\u03ae\u03bc\u03b1\u03c4\u03b1 \u03ad\u03c7\u03bf\u03c5\u03bd \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1: \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 deterministic depth-adaptive recompute \u03b1\u03bd\u03c4\u03af \u03b3\u03b9\u03b1 \u03ad\u03bd\u03b1\u03bd \u03b1\u03bd\u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf predictor. \u03a3\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc AI engineering, \u03b7 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae\u03c2 \u03c5\u03c0\u03bf\u03c7\u03ce\u03c1\u03b7\u03c3\u03b7 \u03c3\u03b5 \u03b3\u03bd\u03c9\u03c3\u03c4\u03ae \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac \u03b1\u03be\u03af\u03b6\u03b5\u03b9 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf \u03b1\u03c0\u03cc \u03bc\u03b9\u03b1 \u03b1\u03c3\u03c4\u03b1\u03b8\u03ae \u03b2\u03b5\u03bb\u03c4\u03b9\u03c3\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03b5\u03af. \u0388\u03bd\u03b1 \u03c3\u03c5\u03b3\u03b3\u03b5\u03bd\u03b9\u03ba\u03cc \u03c0\u03bb\u03b1\u03af\u03c3\u03b9\u03bf \u03b3\u03b9\u03b1 versioned tests, \u03cc\u03c1\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b5\u03c0\u03bf\u03c0\u03c4\u03b5\u03af\u03b1 \u03b4\u03af\u03bd\u03b5\u03b9 \u03b7 <a href=\"https:\/\/twodots.gr\/epalithefsi-politikon-reinforcement-learning-ai-agent\/\">\u03b5\u03c0\u03b1\u03bb\u03ae\u03b8\u03b5\u03c5\u03c3\u03b7 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ce\u03bd \u03b3\u03b9\u03b1 AI agents<\/a>.<\/p>\n<h2 id=\"hardware-boundaries\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf hardware boundary \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c5\u03c0\u03bf\u03c3\u03b7\u03bc\u03b5\u03af\u03c9\u03c3\u03b7<\/h2>\n<p>\u03a4\u03b1 benchmarks \u03ad\u03b3\u03b9\u03bd\u03b1\u03bd \u03c3\u03b5 Apple M4 Max \u03bc\u03b5 64 GB unified memory, Qwen2.5-14B-Instruct Q4_K_M \u03ba\u03b1\u03b9 nomic-embed-text-v1.5. \u03a4\u03bf physical splice \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af \u03ac\u03bc\u03b5\u03c3\u03b7 \u03b1\u03bd\u03c4\u03b9\u03b3\u03c1\u03b1\u03c6\u03ae \u03c3\u03b5 \u03c4\u03bf\u03c0\u03b9\u03ba\u03ad\u03c2 \u03b4\u03b9\u03b5\u03c5\u03b8\u03cd\u03bd\u03c3\u03b5\u03b9\u03c2 cache, \u03b5\u03c0\u03bf\u03bc\u03ad\u03bd\u03c9\u03c2 \u03b1\u03c6\u03bf\u03c1\u03ac UMA-resident \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7. Cloud APIs \u03c0\u03bf\u03c5 \u03b4\u03ad\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf text tokens \u03ba\u03b1\u03b9 discrete-GPU backends \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03bf\u03c5\u03bd \u03c3\u03b5 \u03ba\u03b1\u03bd\u03bf\u03bd\u03b9\u03ba\u03cc text-prefill \u03c7\u03c9\u03c1\u03af\u03c2 acceleration \u03b1\u03c0\u03cc splice.<\/p>\n<p>\u03a4\u03bf Nexus \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 transposed-V layout \u03b3\u03b9\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03bc\u03b5 soft-capped attention \u03cc\u03c0\u03c9\u03c2 \u03c4\u03bf Gemma-2, \u03cc\u03c0\u03bf\u03c5 \u03b1\u03c0\u03b5\u03bd\u03b5\u03c1\u03b3\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 \u03c4\u03bf FlashAttention. \u0397 \u03c5\u03bb\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b8\u03b7\u03ba\u03b5 \u03c9\u03c2 \u03c0\u03c1\u03bf\u03c2 \u03c4\u03b7\u03bd \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c3\u03b5 Gemma-2-9B, \u03b1\u03bb\u03bb\u03ac \u03b1\u03c5\u03c4\u03ae \u03b7 \u03b4\u03bf\u03ba\u03b9\u03bc\u03ae \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03c4\u03b1\u03b9 \u03b5\u03ba\u03c4\u03cc\u03c2 \u03c4\u03bf\u03c5 committed quantitative bundle. \u0394\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bb\u03bf\u03b9\u03c0\u03cc\u03bd \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03b5\u03c1\u03b8\u03bf\u03cd\u03bd \u03bf\u03b9 \u03b1\u03c1\u03b9\u03b8\u03bc\u03bf\u03af \u03c4\u03bf\u03c5 Qwen \u03c3\u03b5 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03ae \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ad\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03ad\u03b1 \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7.<\/p>\n<p>\u0397 \u03c5\u03bb\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 coarse-grained locking \u03b3\u03b9\u03b1 deterministic \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac \u03c3\u03b5 concurrency, cache-line alignment 128 bytes \u03ba\u03b1\u03b9 pool 32 sequence slots \u03b3\u03b9\u03b1 exact-token prefix reuse. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd passing four-thread determinism test, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b9\u03c3\u03c7\u03c5\u03c1\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 throughput gain \u03b1\u03c0\u03cc \u03c4\u03bf allocator hardening.<\/p>\n<aside class=\"td-article-note\">\n<p><strong>\u03a3\u03b1\u03c6\u03ad\u03c2 \u03cc\u03c1\u03b9\u03bf \u03b3\u03b5\u03bd\u03af\u03ba\u03b5\u03c5\u03c3\u03b7\u03c2:<\/strong> \u03cc\u03bb\u03b1 \u03c4\u03b1 headline \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1\u03bd host, \u03ad\u03bd\u03b1 quantized \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf, \u03ad\u03bd\u03b1\u03bd embedder \u03ba\u03b1\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ac end-to-end samples \u03ad\u03c9\u03c2 30 cases. \u03a4\u03bf deep splice \u03bc\u03b5\u03c4\u03c1\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c3\u03b5 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03ba\u03b1\u03b9 latency, \u03cc\u03c7\u03b9 \u03c9\u03c2 \u03c0\u03bb\u03ae\u03c1\u03b7\u03c2 multi-turn \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae.<\/p>\n<\/aside>\n<div class=\"td-decision-band\">\n<div class=\"td-decision-band-content\">\n<p class=\"td-decision-band-kicker\">\u0397 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 AI<\/p>\n<p class=\"td-decision-band-title\">\u03a5\u03b9\u03bf\u03b8\u03b5\u03c4\u03ae\u03c3\u03c4\u03b5 \u03c0\u03c1\u03ce\u03c4\u03b1 \u03c4\u03bf retrieval pattern, \u03cc\u03c7\u03b9 \u03c4\u03bf hardware-specific splice<\/p>\n<p>\u0391\u03bd \u03c4\u03bf registry \u03bc\u03b5\u03b3\u03b1\u03bb\u03ce\u03bd\u03b5\u03b9, \u03be\u03b5\u03ba\u03b9\u03bd\u03ae\u03c3\u03c4\u03b5 \u03bc\u03b5 \u03c3\u03b1\u03c6\u03ae tool names, \u03c3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2, confusion tests \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 fallback. \u0395\u03be\u03b5\u03c4\u03ac\u03c3\u03c4\u03b5 \u03c6\u03c5\u03c3\u03b9\u03ba\u03cc KV splicing \u03bc\u03cc\u03bd\u03bf \u03c3\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf local UMA stack, \u03b1\u03c6\u03bf\u03cd \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b5\u03c4\u03b5 RoPE fidelity, runtime compatibility \u03ba\u03b1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc latency budget.<\/p>\n<\/div>\n<\/div>\n<h2 id=\"nexus-ai-workflows\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03b6\u03bf\u03c5\u03bd AI workflows<\/h2>\n<p>\u0397 \u03c0\u03b9\u03bf \u03ac\u03bc\u03b5\u03c3\u03b1 \u03b1\u03be\u03b9\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03b9\u03bc\u03b7 \u03b1\u03c1\u03c7\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03bc\u03b7\u03bd \u03b5\u03ba\u03c4\u03af\u03b8\u03b5\u03c4\u03b1\u03b9 \u03ba\u03ac\u03b8\u03b5 tool schema \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 turn. \u03a4\u03bf registry \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bf\u03c1\u03b3\u03b1\u03bd\u03c9\u03b8\u03b5\u03af \u03b3\u03b9\u03b1 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7: \u03c3\u03b1\u03c6\u03ae, \u03b4\u03b9\u03b1\u03ba\u03c1\u03b9\u03c4\u03ac \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03b1, \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2 \u03c0\u03bf\u03c5 \u03b5\u03be\u03b7\u03b3\u03bf\u03cd\u03bd \u03c0\u03cc\u03c4\u03b5 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03ba\u03b1\u03b9 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf \u03c3\u03c4\u03ac\u03b4\u03b9\u03bf \u03b3\u03b9\u03b1 \u03b1\u03bc\u03c6\u03af\u03b2\u03bf\u03bb\u03b5\u03c2 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ad\u03c2. \u0388\u03bd\u03b1\u03c2 agent \u03b3\u03b9\u03b1 customer support, \u03b3\u03b9\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1, \u03b4\u03b5\u03bd \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bd\u03b1 \u03b2\u03bb\u03ad\u03c0\u03b5\u03b9 \u03bc\u03b1\u03b6\u03af \u03ba\u03ac\u03b8\u03b5 action \u03c4\u03bf\u03c5 CRM, \u03c4\u03bf\u03c5 ERP \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5 e-shop \u03cc\u03c4\u03b1\u03bd \u03c7\u03b5\u03b9\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 \u03b1\u03c0\u03bb\u03cc \u03b1\u03af\u03c4\u03b7\u03bc\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1\u03c2.<\/p>\n<p>\u0394\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf\u03bd, \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03be\u03b5\u03c7\u03c9\u03c1\u03af\u03b6\u03bf\u03c5\u03bd \u03c4\u03c1\u03af\u03b1 metrics \u03c0\u03bf\u03c5 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c3\u03c5\u03b3\u03c7\u03ad\u03bf\u03bd\u03c4\u03b1\u03b9: routing accuracy, JSON validity \u03ba\u03b1\u03b9 end-to-end argument accuracy. \u0388\u03bd\u03b1\u03c2 agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03ac\u03c8\u03bf\u03b3\u03bf JSON \u03b3\u03b9\u03b1 \u03c4\u03bf \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf. \u0393\u03b9\u03b1 workflows \u03bc\u03b5 CRM, ERP \u03ae publishing, \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03b5\u03b9 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03b1\u03c0\u03cc intent \u03bc\u03ad\u03c7\u03c1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ac \u03c0\u03b5\u03b4\u03af\u03b1 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03b7 \u03c3\u03c5\u03bd\u03c4\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b5\u03b3\u03ba\u03c5\u03c1\u03cc\u03c4\u03b7\u03c4\u03b1. \u0391\u03c5\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03b9\u03b4\u03b9\u03b1\u03af\u03c4\u03b5\u03c1\u03b1 \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03cc \u03c3\u03b5 <a href=\"https:\/\/twodots.gr\/en\/oi-kalyteres-lyseis-help-desk-gia-epicheiriseis\/\">help desk \u03c1\u03bf\u03ad\u03c2 \u03b3\u03b9\u03b1 e-commerce<\/a>, \u03cc\u03c0\u03bf\u03c5 \u03bb\u03ac\u03b8\u03bf\u03c2 routing \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c0\u03b7\u03c1\u03b5\u03ac\u03c3\u03b5\u03b9 \u03c0\u03b5\u03bb\u03ac\u03c4\u03b7 \u03ae \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b1.<\/p>\n<p>\u03a4\u03c1\u03af\u03c4\u03bf\u03bd, \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 fail-closed boundaries. \u0391\u03bd \u03c4\u03bf runtime model, \u03c4\u03bf RoPE configuration \u03ae \u03c4\u03bf cache layout \u03b4\u03b5\u03bd \u03c4\u03b1\u03b9\u03c1\u03b9\u03ac\u03b6\u03b5\u03b9 \u03bc\u03b5 \u03c4\u03bf compiled block, \u03b7 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae\u03c2 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03c1\u03c1\u03b9\u03c6\u03b8\u03b5\u03af \u03c4\u03bf splice \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03c3\u03c4\u03b5\u03af prefill. \u0397 \u03af\u03b4\u03b9\u03b1 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03b9\u03c3\u03c7\u03cd\u03b5\u03b9 \u03c3\u03b5 \u03b5\u03be\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ad\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2: \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 permissions, approval gates \u03b3\u03b9\u03b1 \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b1 writes, idempotency \u03cc\u03c0\u03bf\u03c5 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 trace \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 tool call.<\/p>\n<p>\u03a4\u03ad\u03bb\u03bf\u03c2, \u03c4\u03bf retrieval \u03bc\u03bf\u03bd\u03bf\u03c0\u03ac\u03c4\u03b9 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf \u03b1\u03c0\u03cc speculative system optimizations. \u0391\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03b1\u03bd \u03c4\u03bf KV-cache trick \u03b4\u03b5\u03bd \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf cloud stack, \u03b7 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7 \u03bb\u03af\u03b3\u03c9\u03bd \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd \u03ba\u03b1\u03b9 \u03b7 \u03c3\u03c5\u03bc\u03c0\u03af\u03b5\u03c3\u03b7 schemas \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03bc\u03b5\u03b9\u03ce\u03c3\u03bf\u03c5\u03bd context bloat. \u0397 TWO DOTS \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03b9 \u03b1\u03c5\u03c4\u03ae \u03c4\u03b7 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03c3\u03b5 <a href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03c9\u03bd \u03bc\u03b5 AI<\/a> \u03c0\u03bf\u03c5 \u03be\u03b5\u03ba\u03b9\u03bd\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1, \u03c4\u03b1 \u03b4\u03b9\u03ba\u03b1\u03b9\u03ce\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf fallback \u03b1\u03bd\u03c4\u03af \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 \u03b5\u03bd\u03c4\u03c5\u03c0\u03c9\u03c3\u03b9\u03b1\u03ba\u03cc demo.<\/p>\n<h2 id=\"nexus-evaluation-pilot\">\u03a0\u03ce\u03c2 \u03bd\u03b1 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03ae\u03c3\u03b5\u03c4\u03b5 \u03bc\u03b9\u03b1 \u03c0\u03b1\u03c1\u03cc\u03bc\u03bf\u03b9\u03b1 \u03bb\u03cd\u03c3\u03b7<\/h2>\n<p>\u0388\u03bd\u03b1 \u03c3\u03c9\u03c3\u03c4\u03cc pilot \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03bc\u03b5 registry \u03b1\u03bd\u03c4\u03b9\u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b5\u03c5\u03c4\u03b9\u03ba\u03cc \u03c4\u03bf\u03c5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03bf\u03cd \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd\u03c4\u03bf\u03c2, \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03bc\u03b5 \u03c4\u03b5\u03c7\u03bd\u03b7\u03c4\u03ac \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03b1. \u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 top-1 \u03ba\u03b1\u03b9 top-k recall, \u03b1\u03bc\u03c6\u03af\u03b2\u03bf\u03bb\u03b1 \u03b6\u03b5\u03cd\u03b3\u03b7 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd, cross-turn references \u03ba\u03b1\u03b9 prompts \u03c0\u03bf\u03c5 \u03bc\u03bf\u03b9\u03c1\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c0\u03b1\u03c1\u03cc\u03bc\u03bf\u03b9\u03b1 verbs. \u039c\u03b5\u03c4\u03ac \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 arguments \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b9\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ac routed \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03be\u03b1\u03bd\u03ac end-to-end, \u03ce\u03c3\u03c4\u03b5 \u03bd\u03b1 \u03c6\u03b1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03c0\u03bf\u03cd \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b1 \u03bb\u03ac\u03b8\u03b7.<\/p>\n<p>\u0393\u03b9\u03b1 latency, \u03c7\u03c9\u03c1\u03af\u03c3\u03c4\u03b5 \u03c4\u03bf\u03bd \u03c7\u03c1\u03cc\u03bd\u03bf \u03bc\u03ad\u03c7\u03c1\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf\u03c5, \u03c4\u03bf\u03bd \u03c7\u03c1\u03cc\u03bd\u03bf \u03bc\u03ad\u03c7\u03c1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf argument token \u03ba\u03b1\u03b9 \u03c4\u03bf\u03bd \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03cc \u03c7\u03c1\u03cc\u03bd\u03bf \u03bc\u03ad\u03c7\u03c1\u03b9 \u03ad\u03b3\u03ba\u03c5\u03c1\u03b7 \u03ba\u03bb\u03ae\u03c3\u03b7. \u03a4\u03bf Nexus \u03b1\u03bd\u03ad\u03c6\u03b5\u03c1\u03b5 443,8 ms \u03b3\u03b9\u03b1 \u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf argument token \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 737,3 ms \u03c4\u03bf\u03c5 full-schema oracle, \u03b4\u03b7\u03bb\u03b1\u03b4\u03ae 1,66\u00d7 \u03c4\u03b1\u03c7\u03cd\u03c4\u03b5\u03c1\u03b1, \u03bc\u03b5 \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 80% \u03b5\u03be\u03bf\u03b9\u03ba\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 main-context tokens. \u0391\u03c5\u03c4\u03ac \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c4\u03b7\u03c2 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7\u03c2 \u03c0\u03bb\u03b1\u03c4\u03c6\u03cc\u03c1\u03bc\u03b1\u03c2, \u03cc\u03c7\u03b9 \u03c0\u03c1\u03cc\u03b2\u03bb\u03b5\u03c8\u03b7 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 deployment.<\/p>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf tool-routing pilot<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 1<\/span><strong>\u0391\u03c0\u03bf\u03c4\u03c5\u03c0\u03ce\u03c3\u03c4\u03b5 \u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc registry<\/strong>\n<p>\u039a\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 tool names, \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2, input schemas, \u03b4\u03b9\u03ba\u03b1\u03b9\u03ce\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c3\u03c5\u03c7\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c7\u03c1\u03ae\u03c3\u03b7\u03c2 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c0\u03bf\u03c5 \u03b8\u03b1 \u03c3\u03c5\u03bd\u03b4\u03b5\u03b8\u03bf\u03cd\u03bd.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 2<\/span><strong>\u0394\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03ae\u03c3\u03c4\u03b5 \u03c3\u03cd\u03bd\u03bf\u03bb\u03bf routing \u03c0\u03c1\u03bf\u03b8\u03ad\u03c3\u03b5\u03c9\u03bd<\/strong>\n<p>\u03a3\u03c5\u03bb\u03bb\u03ad\u03be\u03c4\u03b5 \u03b1\u03bb\u03b7\u03b8\u03b9\u03bd\u03ac \u03b1\u03b9\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1, \u03c0\u03b1\u03c1\u03b1\u03c6\u03c1\u03ac\u03c3\u03b5\u03b9\u03c2, \u03b1\u03bc\u03c6\u03af\u03b2\u03bf\u03bb\u03b1 verbs \u03ba\u03b1\u03b9 cross-turn references \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b5\u03ba\u03b8\u03ad\u03c4\u03b5\u03c4\u03b5 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b9\u03ba\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 3<\/span><strong>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 full-schema baseline<\/strong>\n<p>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 context tokens, time to first argument, routing accuracy \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03c0\u03c1\u03b9\u03bd \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c3\u03b5\u03c4\u03b5 retrieval \u03ae \u03c3\u03c5\u03bc\u03c0\u03af\u03b5\u03c3\u03b7.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 4<\/span><strong>\u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 top-1 \u03ba\u03b1\u03b9 top-k retrieval<\/strong>\n<p>\u03a7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae\u03c3\u03c4\u03b5 confusion pairs \u03ba\u03b1\u03b9 \u03c1\u03c5\u03b8\u03bc\u03af\u03c3\u03c4\u03b5 \u03c4\u03bf margin gate \u03ce\u03c3\u03c4\u03b5 \u03bf\u03b9 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b5\u03c2 \u03b1\u03c0\u03bf\u03c6\u03ac\u03c3\u03b5\u03b9\u03c2 \u03bd\u03b1 \u03c0\u03b5\u03c1\u03bd\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 5<\/span><strong>\u03a3\u03c5\u03bc\u03c0\u03b9\u03ad\u03c3\u03c4\u03b5 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03c7\u03b1\u03b8\u03bf\u03cd\u03bd constraints<\/strong>\n<p>\u039a\u03c1\u03b1\u03c4\u03ae\u03c3\u03c4\u03b5 \u03bf\u03bd\u03cc\u03bc\u03b1\u03c4\u03b1 \u03c0\u03b5\u03b4\u03af\u03c9\u03bd, types, required flags \u03ba\u03b1\u03b9 \u03c4\u03b9\u03c2 \u03b5\u03bb\u03ac\u03c7\u03b9\u03c3\u03c4\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03b1\u03c6\u03ad\u03c2 \u03c0\u03bf\u03c5 \u03b1\u03bb\u03bb\u03ac\u03b6\u03bf\u03c5\u03bd \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c3\u03c5\u03bc\u03c0\u03bb\u03ae\u03c1\u03c9\u03c3\u03b7 arguments.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 6<\/span><strong>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 end-to-end \u03b1\u03c3\u03c6\u03ac\u03bb\u03b5\u03b9\u03b1<\/strong>\n<p>\u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf, \u03bb\u03ac\u03b8\u03bf\u03c2 arguments, permissions, retries, idempotency, approval gates \u03ba\u03b1\u03b9 recovery \u03c3\u03b5 \u03b5\u03be\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ac writes.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 7<\/span><strong>\u03a0\u03b5\u03c1\u03ac\u03c3\u03c4\u03b5 \u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf traffic<\/strong>\n<p>\u039e\u03b5\u03ba\u03b9\u03bd\u03ae\u03c3\u03c4\u03b5 \u03bc\u03b5 shadow mode \u03ae \u03c7\u03b1\u03bc\u03b7\u03bb\u03bf\u03cd \u03c1\u03af\u03c3\u03ba\u03bf\u03c5 actions, observability \u03b1\u03bd\u03ac tool call \u03ba\u03b1\u03b9 \u03c0\u03c1\u03bf\u03ba\u03b1\u03b8\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf rollback \u03c3\u03c4\u03bf baseline.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<p>\u0397 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 failure modes \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03bc\u03ad\u03c3\u03bf\u03c5\u03c2 \u03cc\u03c1\u03bf\u03c5\u03c2. \u03a4\u03bf paper \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b5 \u03bc\u03cc\u03bb\u03b9\u03c2 30 cases \u03c3\u03b5 \u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 end-to-end tests \u03ba\u03b1\u03b9 15 trials \u03b1\u03bd\u03ac cell \u03c3\u03c4\u03b1 deep-splice timings. \u039c\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03bf\u03ba\u03b9\u03bc\u03ae \u03bc\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac schemas, payloads, latency, permissions \u03ba\u03b1\u03b9 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b7\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1\u03c2. \u039f \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc\u03c2 \u03bf\u03b4\u03b7\u03b3\u03cc\u03c2 \u03b3\u03b9\u03b1 \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/pos-na-aytomatopoiisete-ergasies-me-techniti-noimosyni\/\">\u03c0\u03ce\u03c2 \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03bc\u03b5 AI<\/a> \u03b2\u03bf\u03b7\u03b8\u03ac \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03b5\u03b8\u03bf\u03cd\u03bd \u03bf\u03b9 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ad\u03c2 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03bc\u03b5 \u03c3\u03b1\u03c6\u03ad\u03c2 process ownership.<\/p>\n<h2 id=\"nexus-main-lesson\">\u0397 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1 \u03c4\u03bf\u03c5 Nexus<\/h2>\n<p>\u03a4\u03bf Nexus \u03b4\u03b5\u03bd \u03bb\u03cd\u03bd\u03b5\u03b9 \u03c4\u03bf scaling \u03c4\u03c9\u03bd AI agents \u03bc\u03b5 \u03ad\u03bd\u03b1 \u03bc\u03b1\u03b3\u03b9\u03ba\u03cc cache trick. \u0391\u03bd\u03b1\u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03cc\u03c4\u03b9 semantic routing, argument generation \u03ba\u03b1\u03b9 low-level cache acceleration \u03ad\u03c7\u03bf\u03c5\u03bd \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03bf\u03cd\u03c2 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2. \u039a\u03c1\u03b1\u03c4\u03ac \u03c4\u03b1 \u03b4\u03cd\u03bf \u03c0\u03c1\u03ce\u03c4\u03b1 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 retrieval \u03ba\u03b1\u03b9 \u03c3\u03c5\u03bc\u03c0\u03b9\u03b5\u03c3\u03bc\u03ad\u03bd\u03bf \u03ba\u03b5\u03af\u03bc\u03b5\u03bd\u03bf, \u03b5\u03bd\u03ce \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03c4\u03bf splicing \u03bc\u03cc\u03bd\u03bf \u03b5\u03ba\u03b5\u03af \u03cc\u03c0\u03bf\u03c5 \u03b7 \u03b8\u03ad\u03c3\u03b7, \u03b7 \u03bc\u03bd\u03ae\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03b7 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03bf\u03cd\u03bd.<\/p>\n<p>\u0391\u03c5\u03c4\u03ae \u03b7 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03b7 \u03c0\u03ad\u03c1\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf prototype. \u038c\u03c3\u03bf \u03c4\u03b1 agent ecosystems \u03b1\u03c0\u03bf\u03ba\u03c4\u03bf\u03cd\u03bd \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 connectors, \u03b7 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 registry, \u03b7 \u03c3\u03c5\u03bc\u03c0\u03af\u03b5\u03c3\u03b7 \u03c4\u03bf\u03c5 interface \u03ba\u03b1\u03b9 \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c4\u03bf\u03c5 routing \u03b3\u03af\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03b2\u03b1\u03c3\u03b9\u03ba\u03ae\u03c2 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae\u03c2. \u03a4\u03bf \u03b6\u03b7\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b4\u03bf\u03b8\u03bf\u03cd\u03bd \u03c3\u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c4\u03b1 \u03c0\u03ac\u03bd\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03bd\u03b1 \u03bb\u03ac\u03b2\u03b5\u03b9 \u03b5\u03b3\u03ba\u03b1\u03af\u03c1\u03c9\u03c2 \u03c4\u03bf \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03b5\u03c0\u03b1\u03c1\u03ba\u03ad\u03c2, \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03b9\u03bc\u03bf interface.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd\u03c4\u03bf\u03c2, \u03b7 \u03ce\u03c1\u03b9\u03bc\u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c4\u03b1\u03b4\u03b9\u03b1\u03ba\u03ae: \u03c0\u03c1\u03ce\u03c4\u03b1 tool taxonomy \u03ba\u03b1\u03b9 retrieval, \u03bc\u03b5\u03c4\u03ac constrained arguments \u03ba\u03b1\u03b9 end-to-end QA, \u03ba\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03cc\u03c4\u03b5 hardware-specific acceleration \u03cc\u03c0\u03bf\u03c5 \u03ad\u03c7\u03b5\u03b9 \u03bc\u03b5\u03c4\u03c1\u03b7\u03bc\u03ad\u03bd\u03bf \u03bd\u03cc\u03b7\u03bc\u03b1. \u03a0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1 \u03b1\u03c0\u03cc <a href=\"https:\/\/twodots.gr\/cars24-ai-agents-exypiretisi-pelaton-leitourgiko-montelo\/\">AI agents \u03c3\u03c4\u03b7\u03bd \u03b5\u03be\u03c5\u03c0\u03b7\u03c1\u03ad\u03c4\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b9\u03c2 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03af\u03b5\u03c2<\/a> \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03b3\u03b9\u03b1\u03c4\u03af \u03b7 \u03c4\u03b5\u03c7\u03bd\u03bf\u03bb\u03bf\u03b3\u03af\u03b1 \u03b1\u03c0\u03bf\u03ba\u03c4\u03ac \u03b1\u03be\u03af\u03b1 \u03bc\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1, \u03c5\u03c0\u03b5\u03cd\u03b8\u03c5\u03bd\u03bf \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1.<\/p>\n<section class=\"td-service-cta\">\n<div class=\"td-service-cta-content\">\n<p class=\"td-service-cta-kicker\">\u0391\u03c0\u03cc \u03c4\u03bf tool registry \u03c3\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7<\/p>\n<p class=\"td-service-cta-title\">\u03a3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 AI workflows \u03c0\u03bf\u03c5 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03bf\u03c5\u03bd \u03c4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03bc\u03b5 \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5\u03c2<\/p>\n<p>\u0397 TWO DOTS \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03b5\u03af integrations, permissions, routing tests, fallbacks \u03ba\u03b1\u03b9 observability, \u03ce\u03c3\u03c4\u03b5 \u03ad\u03bd\u03b1\u03c2 AI agent \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 CRM, ERP, e-shop \u03ae support \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03b1\u03cc\u03c1\u03b1\u03c4\u03bf \u03c1\u03af\u03c3\u03ba\u03bf \u03c3\u03c4\u03b7\u03bd \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u03a3\u03c5\u03b6\u03b7\u03c4\u03ae\u03c3\u03c4\u03b5 \u03c4\u03bf AI workflow \u03c3\u03b1\u03c2<\/a><\/div>\n<\/div>\n<\/section>\n<section id=\"sychnes-erotiseis\" class=\"td-faq-section\">\n<div class=\"td-faq\">\n<p class=\"td-faq-heading\">\u0427\u0435\u0441\u0442\u043e \u0437\u0430\u0434\u0430\u0432\u0430\u043d\u0438 \u0432\u044a\u043f\u0440\u043e\u0441\u0438<\/p>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c0\u03c1\u03bf\u03c3\u03c0\u03b1\u03b8\u03b5\u03af \u03bd\u03b1 \u03bb\u03cd\u03c3\u03b5\u03b9 \u03c4\u03bf Nexus;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039c\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03b1\u03bd\u03cc\u03bc\u03b5\u03bd\u03bf prefill \u03c0\u03bf\u03c5 \u03c0\u03c1\u03bf\u03ba\u03b1\u03bb\u03bf\u03cd\u03bd \u03c4\u03b1 \u03bc\u03b5\u03b3\u03ac\u03bb\u03b1 tool schemas, \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03bf\u03bd\u03c4\u03b1\u03c2 \u03c0\u03c1\u03ce\u03c4\u03b1 \u03c4\u03bf \u03ba\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03bf \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03bf \u03ba\u03b1\u03b9 \u03b4\u03af\u03bd\u03bf\u03bd\u03c4\u03b1\u03c2 \u03c3\u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03b9\u03b1 \u03bc\u03b9\u03ba\u03c1\u03ae \u03c5\u03c0\u03bf\u03b3\u03c1\u03b1\u03c6\u03ae \u03b3\u03b9\u03b1 \u03c4\u03b1 arguments.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0395\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf Nexus \u03ad\u03bd\u03b1 \u03bd\u03ad\u03bf MCP standard;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u0395\u03af\u03bd\u03b1\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03b9\u03ba\u03cc user-space serving prototype \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af MCP-style registries \u03ba\u03b1\u03b9 \u03b5\u03be\u03b5\u03c4\u03ac\u03b6\u03b5\u03b9 retrieval routing, schema compression \u03ba\u03b1\u03b9 KV-cache reuse.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03cc \u03bc\u03bf\u03bd\u03bf\u03c0\u03ac\u03c4\u03b9 \u03c4\u03bf\u03c5;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0388\u03bd\u03b1 INT8 Semantic Lookaside Buffer \u03b1\u03bd\u03b1\u03ba\u03c4\u03ac \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1, \u03ad\u03bd\u03b1 margin gate \u03ba\u03bb\u03b9\u03bc\u03b1\u03ba\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b9\u03c2 \u03b1\u03bc\u03c6\u03af\u03b2\u03bf\u03bb\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03c3\u03b5 cross-encoder \u03ba\u03b1\u03b9 \u03c4\u03b1 arguments \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 compressed textual signature.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0393\u03b9\u03b1\u03c4\u03af \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b7 \u03bc\u03b5\u03c4\u03b1\u03c6\u03bf\u03c1\u03ac \u03c4\u03bf\u03c5 KV cache;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03c0\u03b5\u03b9\u03b4\u03ae \u03c4\u03bf RoPE \u03ba\u03c9\u03b4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03b8\u03ad\u03c3\u03b7. \u0388\u03bd\u03b1 block \u03c0\u03bf\u03c5 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03c4\u03b1\u03b9 \u03b5\u03ba\u03c4\u03cc\u03c2 \u03c4\u03bf\u03c5 \u03b1\u03c1\u03c7\u03b9\u03ba\u03bf\u03cd anchor \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03b9 distributional drift \u03ba\u03b1\u03b9 \u03c3\u03b5 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf \u03b2\u03ac\u03b8\u03bf\u03c2 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 reanchoring \u03ba\u03b1\u03b9 suffix recompute.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u00abnever-regress\u00bb \u03c3\u03c4\u03bf paper;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03c0\u03b9\u03c3\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 output \u03c3\u03b5 \u03c3\u03c7\u03ad\u03c3\u03b7 \u03bc\u03b5 full prefill, \u03bc\u03b5 top-1 agreement \u03ba\u03b1\u03b9 KL divergence \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 \u03bc\u03b7\u03b4\u03ad\u03bd. \u0394\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03bf latency \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u039b\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c4\u03bf physical splicing \u03c3\u03b5 cloud API;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9 \u03c3\u03cd\u03bc\u03c6\u03c9\u03bd\u03b1 \u03bc\u03b5 \u03c4\u03bf scope \u03c4\u03bf\u03c5 paper. \u0391\u03c0\u03b1\u03b9\u03c4\u03b5\u03af \u03ac\u03bc\u03b5\u03c3\u03b7 \u03c0\u03c1\u03cc\u03c3\u03b2\u03b1\u03c3\u03b7 \u03c3\u03b5 \u03c4\u03bf\u03c0\u03b9\u03ba\u03ad\u03c2 physical cache addresses \u03ba\u03b1\u03b9 unified memory\u00b7 \u03c4\u03b1 cloud \u03ba\u03b1\u03b9 discrete-GPU paths \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03bf\u03c5\u03bd \u03c3\u03b5 text-prefill.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u039c\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03bd\u03b1 \u03b3\u03b5\u03bd\u03b9\u03ba\u03b5\u03c5\u03b8\u03bf\u03cd\u03bd \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03ad\u03b1 \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7. \u03a4\u03b1 \u03ba\u03cd\u03c1\u03b9\u03b1 quantitative benchmarks \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1\u03bd M4 Max host \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 Qwen2.5-14B configuration, \u03bc\u03b5 \u03bc\u03b9\u03ba\u03c1\u03ac \u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf \u03bc\u03ac\u03b8\u03b7\u03bc\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03c3\u03b5\u03b9 \u03ac\u03bc\u03b5\u03c3\u03b1 \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd\u03c4\u03bf\u03c2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039d\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 retrieval \u03c0\u03c1\u03b9\u03bd \u03c6\u03bf\u03c1\u03c4\u03ce\u03c3\u03b5\u03b9 schemas, \u03bd\u03b1 \u03ba\u03c1\u03b1\u03c4\u03ac \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03b1\u03bb\u03bb\u03ac \u03c0\u03bb\u03ae\u03c1\u03b5\u03b9\u03c2 tool signatures \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ac routing \u03ba\u03b1\u03b9 arguments end-to-end \u03bc\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 fallback.<\/p>\n<\/div>\n<\/details>\n<\/div>\n<\/section>\n<div class=\"td-source-list\">\n<p id=\"piges\" class=\"td-source-list-title\">\u0418\u0437\u0442\u043e\u0447\u043d\u0438\u0446\u0438<\/p>\n<ul>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2608.20397\" target=\"_blank\" rel=\"noopener\">Arslan \u2014 Nexus: abstract, metadata \u03ba\u03b1\u03b9 scope \u03c3\u03c4\u03bf arXiv<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/html\/2608.20397v1\" target=\"_blank\" rel=\"noopener\">Arslan \u2014 \u03a0\u03bb\u03ae\u03c1\u03b5\u03c2 Nexus paper \u03bc\u03b5 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae, benchmarks \u03ba\u03b1\u03b9 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2<\/a><\/li>\n<li><a href=\"https:\/\/modelcontextprotocol.io\/specification\/2025-11-25\/server\/tools\" target=\"_blank\" rel=\"noopener\">Model Context Protocol \u2014 \u0395\u03c0\u03af\u03c3\u03b7\u03bc\u03b7 \u03c0\u03c1\u03bf\u03b4\u03b9\u03b1\u03b3\u03c1\u03b1\u03c6\u03ae \u03b3\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 JSON schemas<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/1706.03762\" target=\"_blank\" rel=\"noopener\">Vaswani et al. \u2014 Attention Is All You Need<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2104.09864\" target=\"_blank\" rel=\"noopener\">Su et al. \u2014 RoFormer \u03ba\u03b1\u03b9 Rotary Position Embedding<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u03a4\u03bf Nexus \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf MCP tool routing \u03b1\u03c0\u03cc \u03c4\u03bf \u03b2\u03b1\u03c1\u03cd schema-prefill \u03ba\u03b1\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c0\u03bf\u03cd retrieval, compressed signatures \u03ba\u03b1\u03b9 KV-cache splicing \u03ad\u03c7\u03bf\u03c5\u03bd \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1.<\/p>","protected":false},"author":1,"featured_media":97820,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","content-type":"","footnotes":""},"categories":[199],"tags":[6506,20246,20425,19067,20424],"class_list":["post-97149","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-agents","tag-kv-cache","tag-local-llm","tag-mcp","tag-tool-routing"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"_links":{"self":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97149","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/comments?post=97149"}],"version-history":[{"count":1,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97149\/revisions"}],"predecessor-version":[{"id":97821,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97149\/revisions\/97821"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media\/97820"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media?parent=97149"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/categories?post=97149"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/tags?post=97149"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}