{"id":117,"date":"2026-05-15T15:24:51","date_gmt":"2026-05-15T15:24:51","guid":{"rendered":"https:\/\/waspix.com\/index.php\/2026\/05\/15\/hermes-agent-ai-huong-dan-toan-tap-ve-ai-agent-tu-cai-thien-2026\/"},"modified":"2026-05-15T15:24:51","modified_gmt":"2026-05-15T15:24:51","slug":"hermes-agent-ai-huong-dan-toan-tap-ve-ai-agent-tu-cai-thien-2026","status":"publish","type":"post","link":"https:\/\/waspix.com\/index.php\/2026\/05\/15\/hermes-agent-ai-huong-dan-toan-tap-ve-ai-agent-tu-cai-thien-2026\/","title":{"rendered":"Hermes Agent AI: H\u01b0\u1edbng D\u1eabn To\u00e0n T\u1eadp v\u1ec1 AI Agent T\u1ef1 C\u1ea3i Thi\u1ec7n (2026)"},"content":{"rendered":"

1. T\u1ed5ng Quan v\u1ec1 Hermes Agent AI: B\u01b0\u1edbc Ti\u1ebfn \u0110\u1ed9t Ph\u00e1 c\u1ee7a Nous Research<\/h2>\n

Trong k\u1ef7 nguy\u00ean AI 2026, khi c\u00e1c m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef l\u1edbn (LLM) kh\u00f4ng c\u00f2n ch\u1ec9 d\u1eebng l\u1ea1i \u1edf vi\u1ec7c ph\u1ea3n h\u1ed3i v\u0103n b\u1ea3n, s\u1ef1 tr\u1ed7i d\u1eady c\u1ee7a c\u00e1c AI Agent (t\u00e1c nh\u00e2n AI) \u0111\u00e3 thay \u0111\u1ed5i ho\u00e0n to\u00e0n c\u00e1ch ch\u00fang ta l\u00e0m vi\u1ec7c. Hermes Agent AI<\/strong>, m\u1ed9t d\u1ef1 \u00e1n m\u00e3 ngu\u1ed3n m\u1edf t\u1eeb Nous Research<\/strong>, \u0111\u00e3 t\u1ea1o n\u00ean m\u1ed9t c\u01a1n \u0111\u1ecba ch\u1ea5n trong c\u1ed9ng \u0111\u1ed3ng c\u00f4ng ngh\u1ec7 khi \u0111\u1ea1t h\u01a1n 121.000 stars tr\u00ean GitHub ch\u1ec9 trong v\u00f2ng 8 tu\u1ea7n \u0111\u1ea7u ra m\u1eaft \u2014 m\u1ed9t t\u1ed1c \u0111\u1ed9 t\u0103ng tr\u01b0\u1edfng nhanh ch\u01b0a t\u1eebng c\u00f3 \u0111\u1ed1i v\u1edbi m\u1ed9t framework agent.<\/p>\n

1.1. Hermes Agent AI l\u00e0 g\u00ec?<\/h3>\n

Hermes Agent AI kh\u00f4ng ph\u1ea3i l\u00e0 m\u1ed9t chatbot th\u00f4ng th\u01b0\u1eddng hay m\u1ed9t “wrapper” \u0111\u01a1n gi\u1ea3n quanh API c\u1ee7a OpenAI. \u0110\u00e2y l\u00e0 m\u1ed9t h\u1ec7 th\u1ed1ng AI t\u1ef1 v\u1eadn h\u00e0nh (autonomous agent) ho\u00e0n ch\u1ec9nh, \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf \u0111\u1ec3 s\u1ed1ng tr\u00ean server c\u1ee7a b\u1ea1n, ghi nh\u1edb nh\u1eefng g\u00ec n\u00f3 h\u1ecdc \u0111\u01b0\u1ee3c v\u00e0 tr\u1edf n\u00ean th\u00f4ng minh h\u01a1n sau m\u1ed7i t\u00e1c v\u1ee5 th\u1ef1c hi\u1ec7n. \u0110\u01b0\u1ee3c x\u00e2y d\u1ef1ng d\u1ef1a tr\u00ean d\u00f2ng m\u00f4 h\u00ecnh Hermes 4.3 LLM<\/strong>, t\u00e1c nh\u00e2n n\u00e0y t\u1ed1i \u01b0u h\u00f3a kh\u1ea3 n\u0103ng suy lu\u1eadn (reasoning) v\u00e0 s\u1eed d\u1ee5ng c\u00f4ng c\u1ee5 (tool use) \u0111\u1ec3 gi\u1ea3i quy\u1ebft c\u00e1c v\u1ea5n \u0111\u1ec1 ph\u1ee9c t\u1ea1p t\u1eeb nghi\u00ean c\u1ee9u th\u1ecb tr\u01b0\u1eddng \u0111\u1ebfn t\u1ef1 \u0111\u1ed9ng h\u00f3a quy tr\u00ecnh code.<\/p>\n

1.2. Tri\u1ebft l\u00fd “The Agent That Grows With You”<\/h3>\n

H\u1ea7u h\u1ebft c\u00e1c framework agent hi\u1ec7n nay (nh\u01b0 AutoGPT \u0111\u1eddi \u0111\u1ea7u) \u0111\u1ec1u b\u1eaft \u0111\u1ea7u t\u1eeb con s\u1ed1 0 trong m\u1ed7i phi\u00ean l\u00e0m vi\u1ec7c m\u1edbi. Ch\u00fang kh\u00f4ng c\u00f3 k\u00fd \u1ee9c, kh\u00f4ng bi\u1ebft b\u1ea1n th\u00edch \u0111\u1ecbnh d\u1ea1ng b\u00e1o c\u00e1o n\u00e0o, v\u00e0 c\u0169ng kh\u00f4ng nh\u1edb l\u1ea7n tr\u01b0\u1edbc ch\u00fang \u0111\u00e3 g\u1eb7p l\u1ed7i g\u00ec. Hermes Agent AI<\/strong> ph\u00e1 v\u1ee1 r\u00e0o c\u1ea3n n\u00e0y v\u1edbi tri\u1ebft l\u00fd “T\u00e1c nh\u00e2n l\u1edbn l\u00ean c\u00f9ng b\u1ea1n”<\/em>. Th\u00f4ng qua c\u01a1 ch\u1ebf v\u00f2ng l\u1eb7p h\u1ecdc t\u1eadp (learning loop), n\u00f3 t\u00edch l\u0169y ki\u1ebfn th\u1ee9c v\u1ec1 m\u00f4i tr\u01b0\u1eddng l\u00e0m vi\u1ec7c, s\u1edf th\u00edch c\u1ee7a ng\u01b0\u1eddi d\u00f9ng v\u00e0 c\u00e1c quy tr\u00ecnh hi\u1ec7u qu\u1ea3 nh\u1ea5t, gi\u00fap gi\u1ea3m thi\u1ec3u sai s\u1ed1 v\u00e0 ti\u1ebft ki\u1ec7m chi ph\u00ed v\u1eadn h\u00e0nh v\u1ec1 l\u00e2u d\u00e0i.<\/p>\n

2. C\u01a1 Ch\u1ebf Ho\u1ea1t \u0110\u1ed9ng: T\u1ea1i Sao Hermes L\u00e0 “Agent T\u1ef1 C\u1ea3i Thi\u1ec7n”?<\/h2>\n

Kh\u1ea3 n\u0103ng ai agent t\u1ef1 c\u1ea3i thi\u1ec7n<\/strong> c\u1ee7a Hermes kh\u00f4ng ph\u1ea3i l\u00e0 m\u1ed9t l\u1eddi h\u1ee9a marketing, m\u00e0 l\u00e0 m\u1ed9t c\u1ea5u tr\u00fac k\u1ef9 thu\u1eadt d\u1ef1a tr\u00ean ph\u1ea3n h\u1ed3i th\u1ef1c t\u1ebf v\u00e0 l\u01b0u tr\u1eef tri th\u1ee9c b\u1ec1n v\u1eefng.<\/p>\n

2.1. V\u00f2ng l\u1eb7p ReAct v\u00e0 Skill Documents<\/h3>\n

Tr\u00e1i tim c\u1ee7a h\u1ec7 th\u1ed1ng l\u00e0 v\u00f2ng l\u1eb7p ReAct (Reason – Act – Observe). Khi nh\u1eadn m\u1ed9t y\u00eau c\u1ea7u, Hermes s\u1ebd quan s\u00e1t m\u00f4i tr\u01b0\u1eddng, suy lu\u1eadn b\u01b0\u1edbc ti\u1ebfp theo, th\u1ef1c hi\u1ec7n h\u00e0nh \u0111\u1ed9ng (nh\u01b0 g\u1ecdi c\u00f4ng c\u1ee5, truy c\u1eadp web) v\u00e0 quan s\u00e1t k\u1ebft qu\u1ea3. <\/p>\n

\u0110i\u1ec3m kh\u00e1c bi\u1ec7t n\u1eb1m \u1edf giai \u0111o\u1ea1n sau khi ho\u00e0n th\u00e0nh: Skill Documents<\/strong>. Hermes s\u1ebd t\u1ef1 \u0111\u1ed9ng tr\u00edch xu\u1ea5t c\u00e1c quy tr\u00ecnh th\u00e0nh c\u00f4ng, lo\u1ea1i b\u1ecf c\u00e1c chi ti\u1ebft th\u1eeba v\u00e0 \u0111\u00f3ng g\u00f3i ch\u00fang th\u00e0nh c\u00e1c t\u1ec7p Markdown trong th\u01b0 m\u1ee5c k\u1ef9 n\u0103ng. Theo c\u00e1c benchmark n\u1ed9i b\u1ed9 c\u1ee7a Nous Research, vi\u1ec7c s\u1eed d\u1ee5ng l\u1ea1i Skill Documents gi\u00fap gi\u1ea3m t\u1edbi 40% th\u1eddi gian th\u1ef1c hi\u1ec7n t\u00e1c v\u1ee5 so v\u1edbi m\u1ed9t agent m\u1edbi ho\u00e0n to\u00e0n, do n\u00f3 kh\u00f4ng ph\u1ea3i “m\u00f2 m\u1eabm” l\u1ea1i c\u00e1c b\u01b0\u1edbc kh\u00e1m ph\u00e1 c\u00f4ng c\u1ee5.<\/p>\n

2.2. H\u1ec7 th\u1ed1ng b\u1ed9 nh\u1edb ba t\u1ea7ng (Long-term Memory)<\/h3>\n

B\u1ed9 nh\u1edb ai agent<\/strong> c\u1ee7a Hermes \u0111\u01b0\u1ee3c t\u1ed5 ch\u1ee9c ch\u1eb7t ch\u1ebd \u0111\u1ec3 \u0111\u1ea3m b\u1ea3o t\u00ednh nh\u1ea5t qu\u00e1n qua nhi\u1ec1u tu\u1ea7n, th\u1eadm ch\u00ed nhi\u1ec1u th\u00e1ng:<\/p>\n

    \n
  • MEMORY.md<\/code>: L\u01b0u tr\u1eef ng\u1eef c\u1ea3nh ng\u1eafn h\u1ea1n, m\u1ee5c ti\u00eau t\u1ee9c th\u1eddi v\u00e0 tr\u1ea1ng th\u00e1i c\u1ee7a t\u00e1c v\u1ee5 \u0111ang ch\u1ea1y.<\/li>\n
  • USER.md<\/code>: M\u1ed9t kho l\u01b0u tr\u1eef d\u00e0i h\u1ea1n v\u1ec1 s\u1edf th\u00edch ng\u01b0\u1eddi d\u00f9ng, c\u00e1c quy chu\u1ea9n v\u1ec1 ch\u1ea5t l\u01b0\u1ee3ng v\u00e0 ghi ch\u00fa c\u00e1 nh\u00e2n h\u00f3a.<\/li>\n
  • SQLite Database:<\/strong> L\u01b0u tr\u1eef to\u00e0n b\u1ed9 l\u1ecbch s\u1eed h\u1ed9i tho\u1ea1i v\u00e0 c\u00e1c phi\u00ean l\u00e0m vi\u1ec7c c\u0169, cho ph\u00e9p agent truy h\u1ed3i th\u00f4ng tin ch\u00ednh x\u00e1c th\u00f4ng qua t\u00ecm ki\u1ebfm ng\u1eef ngh\u0129a (semantic search).<\/li>\n<\/ul>\n

    3. H\u01b0\u1edbng D\u1eabn C\u00e0i \u0110\u1eb7t Hermes Agent Chi Ti\u1ebft<\/h2>\n

    Vi\u1ec7c tri\u1ec3n khai Hermes Agent \u0111\u00e3 \u0111\u01b0\u1ee3c \u0111\u01a1n gi\u1ea3n h\u00f3a t\u1ed1i \u0111a \u0111\u1ec3 b\u1ea5t k\u1ef3 ai c\u0169ng c\u00f3 th\u1ec3 th\u1ef1c hi\u1ec7n tr\u00ean Linux, macOS ho\u1eb7c Windows (qua WSL2).<\/p>\n

    3.1. C\u00e0i \u0111\u1eb7t nhanh qua Script Shell<\/h3>\n

    \u0110\u00e2y l\u00e0 ph\u01b0\u01a1ng th\u1ee9c nhanh nh\u1ea5t \u0111\u01b0\u1ee3c khuy\u00ean d\u00f9ng trong c\u00e1c b\u1ea3n release m\u1edbi nh\u1ea5t (v0.11.0):<\/p>\n

    curl -fsSL https:\/\/hermes-agent.nousresearch.com\/install.sh | bash<\/code><\/p>\n

    L\u1ec7nh n\u00e0y s\u1ebd t\u1ef1 \u0111\u1ed9ng thi\u1ebft l\u1eadp m\u00f4i tr\u01b0\u1eddng, ki\u1ec3m tra c\u00e1c ph\u1ee5 thu\u1ed9c v\u00e0 c\u00e0i \u0111\u1eb7t c\u00f4ng c\u1ee5 CLI ch\u00ednh th\u1ee9c.<\/p>\n

    3.2. C\u00e0i \u0111\u1eb7t qua Docker v\u00e0 c\u00f4ng c\u1ee5 `uv`<\/h3>\n

    \u0110\u1ed1i v\u1edbi c\u00e1c developer mu\u1ed1n s\u1ef1 \u1ed5n \u0111\u1ecbnh v\u00e0 c\u00f4 l\u1eadp ho\u00e0n to\u00e0n, Docker l\u00e0 l\u1ef1a ch\u1ecdn h\u00e0ng \u0111\u1ea7u. N\u1ebfu b\u1ea1n mu\u1ed1n c\u00e0i \u0111\u1eb7t th\u1ee7 c\u00f4ng \u0111\u1ec3 \u0111\u00f3ng g\u00f3p (contribute) cho d\u1ef1 \u00e1n, h\u00e3y s\u1eed d\u1ee5ng uv<\/code> \u2014 tr\u00ecnh qu\u1ea3n l\u00fd package c\u1ef1c nhanh c\u1ee7a Astral:<\/p>\n

    git clone https:\/\/github.com\/NousResearch\/hermes-agent.git\ncd hermes-agent\ncurl -LsSf https:\/\/astral.sh\/uv\/install.sh | sh\nuv venv venv --python 3.11\nsource venv\/bin\/activate\nuv pip install -e \".[all,dev]\"<\/code><\/pre>\n
    3.3. H\u01b0\u1edbng d\u1eabn s\u1eed d\u1ee5ng Hermes Agent CLI<\/h3>\n
    Sau khi c\u00e0i \u0111\u1eb7t, b\u1ea1n s\u1ebd t\u01b0\u01a1ng t\u00e1c ch\u1ee7 y\u1ebfu qua d\u00f2ng l\u1ec7nh. M\u1ed9t s\u1ed1 l\u1ec7nh h\u01b0\u1edbng d\u1eabn hermes agent cli<\/strong> quan tr\u1ecdng bao g\u1ed3m:<\/p>\n
      \n
    • hermes setup<\/code>: Ch\u1ea1y tr\u00ecnh h\u01b0\u1edbng d\u1eabn \u0111\u1ec3 c\u1ea5u h\u00ecnh API Key (OpenAI, Anthropic ho\u1eb7c Nous Portal).<\/li>\n
    • hermes model<\/code>: Chuy\u1ec3n \u0111\u1ed5i gi\u1eefa c\u00e1c m\u00f4 h\u00ecnh LLM kh\u00e1c nhau m\u1ed9t c\u00e1ch nhanh ch\u00f3ng.<\/li>\n
    • hermes doctor<\/code>: Ki\u1ec3m tra tr\u1ea1ng th\u00e1i h\u1ec7 th\u1ed1ng v\u00e0 k\u1ebft n\u1ed1i Docker backend.<\/li>\n
    • hermes --continue<\/code>: Ti\u1ebfp t\u1ee5c phi\u00ean l\u00e0m vi\u1ec7c c\u0169 m\u00e0 kh\u00f4ng l\u00e0m m\u1ea5t ng\u1eef c\u1ea3nh.<\/li>\n
    • S\u1eed d\u1ee5ng l\u1ec7nh \/save<\/code> ngay trong phi\u00ean chat \u0111\u1ec3 l\u01b0u l\u1ea1i c\u00e1c k\u1ef9 n\u0103ng v\u1eeba h\u1ecdc.<\/li>\n<\/ul>\n
      4. L\u1ef1a Ch\u1ecdn “B\u1ed9 N\u00e3o”: Hermes 4.3 LLM v\u00e0 Ollama<\/h2>\n
      Hi\u1ec7u su\u1ea5t c\u1ee7a agent ph\u1ee5 thu\u1ed9c r\u1ea5t l\u1edbn v\u00e0o m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef (LLM) \u0111\u1ee9ng sau n\u00f3.<\/p>\n
      4.1. S\u1ee9c m\u1ea1nh c\u1ee7a Hermes 4.3 36B<\/h3>\n
      D\u00f2ng m\u00f4 h\u00ecnh Hermes 4.3 LLM<\/strong> (\u0111\u1eb7c bi\u1ec7t l\u00e0 b\u1ea3n 36B) \u0111\u01b0\u1ee3c \u0111\u00e0o t\u1ea1o th\u00f4ng qua khung Atropos<\/em>, s\u1eed d\u1ee5ng h\u00e0ng ng\u00e0n tr\u00ecnh x\u00e1c minh t\u00e1c v\u1ee5 (task verifiers) \u0111\u1ec3 c\u1ee7ng c\u1ed1 c\u00e1c \u0111\u1ea7u ra \u0111\u00fang. \u0110i\u1ec1u n\u00e0y khi\u1ebfn n\u00f3 v\u01b0\u1ee3t tr\u1ed9i h\u01a1n c\u00e1c m\u00f4 h\u00ecnh generalist kh\u00e1c trong vi\u1ec7c hi\u1ec3u \u0111\u1ecbnh d\u1ea1ng <tool_call><\/code> v\u00e0 th\u1ef1c thi c\u00e1c chu\u1ed7i l\u1ec7nh logic ph\u1ee9c t\u1ea1p.<\/p>\n
      4.2. K\u1ebft n\u1ed1i v\u1edbi LLM c\u1ee5c b\u1ed9 (Ollama)<\/h3>\n
      N\u1ebfu b\u1ea1n quan t\u00e2m \u0111\u1ebfn quy\u1ec1n ri\u00eang t\u01b0 ho\u1eb7c mu\u1ed1n c\u1eaft gi\u1ea3m chi ph\u00ed API, vi\u1ec7c k\u1ebft n\u1ed1i v\u1edbi Ollama<\/strong> l\u00e0 gi\u1ea3i ph\u00e1p ho\u00e0n h\u1ea3o. Hermes Agent h\u1ed7 tr\u1ee3 \u0111\u1ea7y \u0111\u1ee7 c\u00e1c endpoint t\u01b0\u01a1ng th\u00edch OpenAI. Ch\u1ec9 c\u1ea7n ch\u1ea1y Ollama c\u1ee5c b\u1ed9 v\u00e0 tr\u1ecf endpoint t\u1edbi http:\/\/localhost:11434\/v1<\/code> trong qu\u00e1 tr\u00ecnh hermes setup<\/code>. L\u01b0u \u00fd: \u0110\u1ec3 ch\u1ea1y m\u00f4 h\u00ecnh 36B m\u01b0\u1ee3t m\u00e0, b\u1ea1n n\u00ean c\u00f3 GPU v\u1edbi VRAM t\u1ed1i thi\u1ec3u 24GB.<\/p>\n
      5. So S\u00e1nh Hermes Agent vs CrewAI vs LangGraph<\/h2>\n
      Vi\u1ec7c l\u1ef1a ch\u1ecdn framework ph\u00f9 h\u1ee3p v\u1edbi workflow l\u00e0 c\u1ef1c k\u1ef3 quan tr\u1ecdng. D\u01b0\u1edbi \u0111\u00e2y l\u00e0 b\u1ea3ng so s\u00e1nh chi ti\u1ebft:<\/p>\n\n\n\n\n\n\n\n\n
      Ti\u00eau ch\u00ed<\/th>\nHermes Agent AI<\/th>\nCrewAI<\/th>\nLangGraph<\/th>\n<\/tr>\n<\/thead>\n
      C\u01a1 ch\u1ebf b\u1ed9 nh\u1edb<\/strong><\/td>\nT\u1ef1 \u0111\u1ed9ng (3 t\u1ea7ng), l\u01b0u tr\u1eef v\u0129nh vi\u1ec5n<\/td>\nTh\u1ee7 c\u00f4ng th\u00f4ng qua Shared Memory<\/td>\nQu\u1ea3n l\u00fd qua State Graph ph\u1ee9c t\u1ea1p<\/td>\n<\/tr>\n
      Kh\u1ea3 n\u0103ng t\u1ef1 h\u1ecdc<\/strong><\/td>\nC\u00f3 (T\u1ef1 t\u1ea1o Skill Documents)<\/td>\nKh\u00f4ng (D\u1ef1a tr\u00ean Role-play)<\/td>\nKh\u00f4ng (C\u1ea7n \u0111\u1ecbnh ngh\u0129a l\u1ea1i lu\u1ed3ng)<\/td>\n<\/tr>\n
      \u0110\u1ed9 kh\u00f3 setup<\/strong><\/td>\nD\u1ec5 (5-10 ph\u00fat)<\/td>\nTrung b\u00ecnh (C\u1ea7n code Python)<\/td>\nCao (D\u00e0nh cho k\u1ef9 s\u01b0 h\u1ec7 th\u1ed1ng)<\/td>\n<\/tr>\n
      \u1ee8ng d\u1ee5ng t\u1ed1t nh\u1ea5t<\/strong><\/td>\nTr\u1ee3 l\u00fd c\u00e1 nh\u00e2n, t\u1ef1 \u0111\u1ed9ng h\u00f3a l\u1eb7p l\u1ea1i<\/td>\nPh\u1ed1i h\u1ee3p nhi\u1ec1u vai tr\u00f2 agent<\/td>\nQuy tr\u00ecnh doanh nghi\u1ec7p nghi\u00eam ng\u1eb7t<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
      So s\u00e1nh hermes agent vs crewai<\/strong> cho th\u1ea5y: CrewAI m\u1ea1nh v\u1ec1 vi\u1ec7c chia vi\u1ec7c cho m\u1ed9t “nh\u00f3m” nh\u00e2n vi\u00ean AI, trong khi Hermes t\u1eadp trung v\u00e0o vi\u1ec7c bi\u1ebfn m\u1ed9t “nh\u00e2n vi\u00ean” duy nh\u1ea5t tr\u1edf n\u00ean xu\u1ea5t ch\u00fang qua th\u1eddi gian.<\/p>\n
      6. \u0110\u00e1nh Gi\u00e1 Chi Ph\u00ed V\u1eadn H\u00e0nh v\u00e0 ROI<\/h2>\n
      M\u1ed9t c\u00e2u h\u1ecfi l\u1edbn c\u1ee7a ng\u01b0\u1eddi d\u00f9ng l\u00e0: “T\u1ed1n bao nhi\u00eau ti\u1ec1n \u0111\u1ec3 ch\u1ea1y Hermes?”. <\/p>\n
        \n
      • Chi ph\u00ed API:<\/strong> M\u1ed9t t\u00e1c v\u1ee5 nghi\u00ean c\u1ee9u th\u1ecb tr\u01b0\u1eddng \u0111i\u1ec3n h\u00ecnh ti\u00eau t\u1ed1n kho\u1ea3ng 15.000 – 25.000 tokens. V\u1edbi m\u00f4 h\u00ecnh Hermes 4.3 36B qua Nous Portal, chi ph\u00ed ch\u1ec9 r\u01a1i v\u00e0o kho\u1ea3ng $0.03 – $0.05 m\u1ed7i t\u00e1c v\u1ee5.<\/li>\n
      • Kh\u1ea3 n\u0103ng ti\u1ebft ki\u1ec7m:<\/strong> Nh\u1edd c\u01a1 ch\u1ebf Skill Documents, c\u00e1c l\u1ea7n ch\u1ea1y sau cho c\u00f9ng m\u1ed9t lo\u1ea1i t\u00e1c v\u1ee5 s\u1ebd t\u1ed1n \u00edt token h\u01a1n \u0111\u00e1ng k\u1ec3 do agent b\u1ecf qua \u0111\u01b0\u1ee3c giai \u0111o\u1ea1n “th\u1eed sai”.<\/li>\n
      • M\u00f4 h\u00ecnh c\u1ee5c b\u1ed9:<\/strong> Ch\u1ea1y qua Ollama gi\u00fap chi ph\u00ed API v\u1ec1 0, nh\u01b0ng b\u1ea1n c\u1ea7n \u0111\u1ea7u t\u01b0 ph\u1ea7n c\u1ee9ng (GPU) ban \u0111\u1ea7u.<\/li>\n<\/ul>\n
        7. \u01afu \u0110i\u1ec3m v\u00e0 Th\u00e1ch Th\u1ee9c Khi Tri\u1ec3n Khai<\/h2>\n
        7.1. \u01afu \u0111i\u1ec3m<\/h3>\n
          \n
        • S\u1ef1 b\u1ec1n b\u1ec9 c\u1ee7a tri th\u1ee9c:<\/strong> \u0110\u00e2y l\u00e0 framework hi\u1ebfm hoi th\u1ef1c s\u1ef1 gi\u1ea3i quy\u1ebft \u0111\u01b0\u1ee3c b\u00e0i to\u00e1n “m\u1ea5t tr\u00ed nh\u1edb” c\u1ee7a AI agent.<\/li>\n
        • C\u1ed9ng \u0111\u1ed3ng m\u1ea1nh m\u1ebd:<\/strong> V\u1edbi h\u00e0ng ng\u00e0n k\u1ef9 n\u0103ng \u0111\u01b0\u1ee3c chia s\u1ebb t\u1ea1i agentskills.io<\/code>, b\u1ea1n c\u00f3 th\u1ec3 t\u1ea3i v\u1ec1 c\u00e1c quy tr\u00ecnh c\u00f3 s\u1eb5n thay v\u00ec t\u1ef1 x\u00e2y d\u1ef1ng t\u1eeb \u0111\u1ea7u.<\/li>\n
        • T\u00ednh linh ho\u1ea1t:<\/strong> H\u1ed7 tr\u1ee3 t\u1eeb cloud (Hermes Atlas) \u0111\u1ebfn local hosting.<\/li>\n<\/ul>\n
          7.2. Th\u00e1ch th\u1ee9c (G\u00f3c nh\u00ecn Senior Auditor)<\/h3>\n
            \n
          • Skill Drift (Sai l\u1ec7ch k\u1ef9 n\u0103ng):<\/strong> N\u1ebfu l\u1ea7n \u0111\u1ea7u agent h\u1ecdc sai quy tr\u00ecnh nh\u01b0ng v\u1eabn l\u01b0u th\u00e0nh Skill, n\u00f3 s\u1ebd l\u1eb7p l\u1ea1i l\u1ed7i \u0111\u00f3. C\u1ea7n c\u00f3 s\u1ef1 gi\u00e1m s\u00e1t c\u1ee7a con ng\u01b0\u1eddi trong giai \u0111o\u1ea1n \u0111\u1ea7u.<\/li>\n
          • B\u1ea3o m\u1eadt d\u1eef li\u1ec7u:<\/strong> Vi\u1ec7c l\u01b0u tr\u1eef b\u1ed9 nh\u1edb v\u00e0o file Markdown (USER.md) gi\u00fap d\u1ec5 \u0111\u1ecdc nh\u01b0ng c\u1ea7n \u0111\u01b0\u1ee3c b\u1ea3o v\u1ec7 c\u1ea9n th\u1eadn n\u1ebfu ch\u1ee9a d\u1eef li\u1ec7u nh\u1ea1y c\u1ea3m c\u1ee7a kh\u00e1ch h\u00e0ng.<\/li>\n<\/ul>\n
            8. K\u1ebft Lu\u1eadn: T\u01b0\u01a1ng Lai c\u1ee7a T\u1ef1 \u0110\u1ed9ng H\u00f3a v\u1edbi Hermes<\/h2>\n
            Hermes Agent AI<\/strong> kh\u00f4ng ch\u1ec9 l\u00e0 m\u1ed9t c\u00f4ng c\u1ee5, n\u00f3 \u0111\u1ea1i di\u1ec7n cho m\u1ed9t th\u1ebf h\u1ec7 “c\u00f4ng nh\u00e2n k\u1ef9 thu\u1eadt s\u1ed1” c\u00f3 kh\u1ea3 n\u0103ng h\u1ecdc h\u1ecfi. Kh\u1ea3 n\u0103ng t\u1ef1 t\u1ea1o Skill Documents l\u00e0 ch\u00eca kh\u00f3a \u0111\u1ec3 \u0111\u01b0a AI t\u1eeb m\u1ed9t chatbot tr\u1ea3 l\u1eddi c\u00e2u h\u1ecfi th\u00e0nh m\u1ed9t tr\u1ee3 l\u00fd th\u1ef1c th\u1ee5 c\u00f3 kh\u1ea3 n\u0103ng t\u1ed1i \u01b0u h\u00f3a hi\u1ec7u su\u1ea5t theo c\u1ea5p s\u1ed1 nh\u00e2n.<\/p>\n
            N\u1ebfu b\u1ea1n l\u00e0 m\u1ed9t c\u00e1 nh\u00e2n mu\u1ed1n t\u1ed1i \u01b0u h\u00f3a workflow h\u00e0ng ng\u00e0y, ho\u1eb7c m\u1ed9t agency mu\u1ed1n x\u00e2y d\u1ef1ng h\u1ec7 th\u1ed1ng t\u1ef1 \u0111\u1ed9ng h\u00f3a n\u1ed9i dung b\u1ec1n v\u1eefng, Hermes Agent AI l\u00e0 \u0111i\u1ec3m xu\u1ea5t ph\u00e1t kh\u00f4ng th\u1ec3 b\u1ecf qua trong n\u0103m 2026. H\u00e3y b\u1eaft \u0111\u1ea7u t\u1eeb nh\u1eefng t\u00e1c v\u1ee5 nh\u1ecf, ki\u00ean nh\u1eabn d\u1ea1y cho agent c\u1ee7a b\u1ea1n nh\u1eefng “ng\u00f3n ngh\u1ec1” \u0111\u1ea7u ti\u00ean, v\u00e0 b\u1ea1n s\u1ebd th\u1ea5y th\u00e0nh qu\u1ea3 sau v\u00e0i tu\u1ea7n s\u1eed d\u1ee5ng.<\/p>\n","protected":false},"excerpt":{"rendered":"
            Kh\u00e1m ph\u00e1 Hermes Agent AI t\u1eeb Nous Research. H\u01b0\u1edbng d\u1eabn chi ti\u1ebft c\u00e1ch c\u00e0i \u0111\u1eb7t, c\u01a1 ch\u1ebf t\u1ef1 h\u1ecdc qua Skill Documents v\u00e0 so s\u00e1nh v\u1edbi CrewAI, LangGraph. Xem ngay!<\/p>\n","protected":false},"author":5,"featured_media":115,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-117","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/posts\/117","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/comments?post=117"}],"version-history":[{"count":0,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/posts\/117\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/media\/115"}],"wp:attachment":[{"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/media?parent=117"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/categories?post=117"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/waspix.com\/index.php\/wp-json\/wp\/v2\/tags?post=117"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}