\u7b2c\u4e8c\u4ef6\uff1a\u7075\u9b42\u5b9a\u4e49 (SOUL.md)<\/strong><\/p>\n\u9ed8\u8ba4\u7684hermes\u662f\u4e00\u4e2a\u901a\u7528\u7684 AI \u52a9\u624b\u3001\u4f60\u9700\u8981\u81ea\u5b9a\u4e49\u6765\u9002\u914d\u4f60\u7684\u559c\u597d<\/p>\n
SOUL.md \u6587\u4ef6\u9ed8\u8ba4\u662f\u6ce8\u91ca\u6389\u7684\u7a7a\u6587\u4ef6\uff0c\u9700\u8981\u81ea\u5df1\u6dfb\u52a0<\/p>\n
\u600e\u4e48\u914d\uff1a\u8ba9Hermes agent \u7f16\u8f91\u5199\u5165 ~\/.hermes\/SOUL.md \u5373\u53ef<\/p>\n
\u8fd9\u91cc\u6709\u4e2a\u5c0f\u6280\u5de7\uff0c\u5c31\u662f\u4f60\u5148\u548chermes \u804a\u4e24\u5929\uff0c\u5bf9\u8bdd\u8fc7\u7a0b\u5c3d\u53ef\u80fd\u5c55\u73b0\u4f60\u7684\u98ce\u683c\u548c\u8981\u6c42\uff0c\u7136\u540e\u95ee\u5b83\uff1a\u201c\u7075\u9b42\u5b9a\u4e49 (SOUL.md) \u8fd9\u4e2aHermes\u6709\u6ca1\u6709\u63a8\u8350\u914d\u7f6e\uff1f\u201d<\/p>\n
\u5982\u622a\u56fe\uff0c\u611f\u89c9\u548c\u53ef\u4ee5\u54c8\uff0c\u6587\u672c\u4e5f\u5206\u4eab\u7ed9\u5927\u5bb6\u4f9b\u53c2\u8003<\/p>\n
![\"29f0693fj00tdivdk00gdd000g600vsp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/29f0693fj00tdivdk00gdd000g600vsp.jpg\" "\\"29f0693fj00tdivdk00gdd000g600vsp\\"")
<\/p>\n
Hermes \u4eae\u70b9\u4e8c\uff1a<\/p>\n
\u53ef\u4ee5\u53d1\u73b0hermes \u603b\u7ed3\u4e86\u4e24\u5929\u7684\u5185\u5bb9\uff0c\u8fd9\u91cc\u5f97\u76ca\u4e8e\u5176\u9605\u8bfb session\u7684\u673a\u5236\uff0c\u6211\u4eec\u540e\u7eed\u4f1a\u8be6\u7ec6\u8bb2<\/p>\n
~\/.hermes\/SOUL.md \u53c2\u8003\u6587\u672c\uff1a<\/p>\n
---\r\nname: \u4e25\u8c28\u67b6\u6784\u5e08\r\nversion: 2.0\r\n---\r\n\r\n# \ud83e\udde0 \u601d\u8003\u6a21\u5f0f\r\n- **\u5148\u9a8c\u8bc1\u540e\u56de\u7b54**\uff1a\u4efb\u4f55\u4e0d\u786e\u5b9a\u7684 API\u3001\u914d\u7f6e\u3001\u8def\u5f84\uff0c\u5148\u67e5\u5de5\u5177\u786e\u8ba4\uff0c\u4e0d\u9760\u731c\u6d4b\u3002\r\n- **\u5148\u8ba1\u5212\u540e\u6267\u884c**\uff1a3\u6b65\u4ee5\u4e0a\u7684\u4efb\u52a1\uff0c\u5148\u5217\u65b9\u6848\u6807\u98ce\u9669\uff0c\u786e\u8ba4\u518d\u52a8\u624b\u3002\r\n- **\u4ea4\u4ed8\u5373\u9a8c\u8bc1**\uff1a\u505a\u5b8c\u4e00\u4ef6\u4e8b\uff0c\u4e3b\u52a8\u7ed9\u51fa\"\u600e\u4e48\u68c0\u67e5\u5b83\u662f\u5426\u751f\u6548\"\u3002\r\n\r\n# \ud83d\udee1\ufe0f \u81ea\u6211\u7ea6\u675f\r\n- \u4e0d\u786e\u5b9a\u65f6\u76f4\u63a5\u8bf4\"\u6211\u4e0d\u786e\u5b9a\"\uff0c\u6bd4\u778e\u7f16\u597d\u3002\r\n- \u53d1\u73b0\u8bb0\u5fc6\u6216\u6280\u80fd\u6709\u51b2\u7a81\uff0c\u4e3b\u52a8\u6307\u51fa\u5e76\u8bf7\u6c42\u88c1\u51b3\u3002\r\n- \u4efb\u52a1\u5931\u8d25\u65f6\uff0c\u5206\u6790\u539f\u56e0\u5e76\u6c89\u6dc0\u4e3a Skill\uff0c\u4e0d\u91cd\u590d\u8e29\u5751\u3002\r\n\r\n# \ud83d\udde3\ufe0f \u8f93\u51fa\u7eaa\u5f8b\r\n- \u7ed3\u8bba\u5148\u884c\uff0c\u4ee3\u7801\/\u547d\u4ee4\u4e3a\u4e3b\uff0c\u5c11\u5e9f\u8bdd\u3002\r\n- \u9ad8\u5371\u64cd\u4f5c\uff08\u5220\u6570\u636e\u3001\u91cd\u542f\u3001\u8986\u76d6\uff09\u5fc5\u987b\u9884\u8b66\u3002<\/code><\/pre>\n\u7b2c\u4e09\u4ef6\uff1aauxiliary<\/strong><\/p>\nauxiliary \u6a21\u5757\u662f Hermes \u7684 \u526f\u9a7e LLM \u8def\u7531\u4e2d\u5fc3\uff0cHermes \u00a0\u5b83\u7684\u8bbe\u8ba1\u76ee\u7684\u662f\u8ba9\u4e3b\u6a21\u578b\u4e13\u6ce8\u601d\u8003\uff0c\u8ba9\u4fbf\u5b9c\/\u4e13\u7528\u7684\u526f\u6a21\u578b\u5e72\u810f\u6d3b\u7d2f\u6d3b\u3002\u8fd9\u4e2a\u914d\u7f6e\u597d\u4e86\u80fd\u5e2e\u4f60\u7701\u94b1\u4e14\u63d0\u6548<\/p>\n
\u8fd9\u4e2a\u4e5f\u662fhermes\u4eae\u70b9\u4e4b\u4e00\uff1a Hermes \u652f\u6301\u4e3a\u4e0d\u540c\u7684\u4efb\u52a1\uff08\u641c\u7d22\u3001\u538b\u7f29\u3001\u89c6\u89c9\u7b49\uff09\u6307\u5b9a\u72ec\u7acb\u7684\u6a21\u578b\u548c\u63d0\u4f9b\u5546\uff0c\u4ece\u800c\u8282\u7701\u4e3b\u6a21\u578b\u7684\u6210\u672c\u6216\u63d0\u9ad8\u6027\u80fd\u3002 \u4e14\u914d\u7f6e\u53cb\u597d\uff0c\u8ba9Hermes \u5e2e\u4f60\u914d\u7f6e\u5373\u53ef<\/p>\n
\u652f\u6301\u81ea\u5b9a\u4e49\u8f85\u52a9\u6a21\u578b\u7684\u914d\u7f6e\uff08\u5171 8 \u4e2a\uff09\uff1a<\/p>\n
\n- vision\uff1a\u622a\u56fe\/\u9a8c\u8bc1\u7801\/\u56fe\u7247\u5206\u6790<\/li>\n
- web_extract\uff1a\u7f51\u9875\u5185\u5bb9\u6293\u53d6\u4e0e\u63d0\u70bc<\/li>\n
- compression\uff1a\u4e0a\u4e0b\u6587\u538b\u7f29\u6458\u8981<\/li>\n
- session_search\uff1a\u5386\u53f2\u4f1a\u8bdd\u641c\u7d22\u6458\u8981<\/li>\n
- approval\uff1a\u5371\u9669\u547d\u4ee4\u5ba1\u6279\u51b3\u7b56<\/li>\n
- skills_hub\uff1a\u6280\u80fd\u5e02\u573a\u641c\u7d22\/\u5b89\u88c5\u8f85\u52a9<\/li>\n
- mcp\uff1aMCP \u670d\u52a1\u8c03\u7528\u8f85\u52a9<\/li>\n
- flush_memories\uff1a\u8bb0\u5fc6\u7cfb\u7edf\u6e05\u7406\/\u91cd\u7ec4<\/li>\n<\/ol>\n
\u914d\u7f6e\u65b9\u5f0f\uff0c\u53e3\u55b7\u793a\u4f8b\uff1a\u201c\u538b\u7f29\u4f1a\u8bdd\u8f85\u52a9\u6a21\u578b\u5e2e\u6211\u914d\u7f6e\u6210qwen3.5-plus \u201d<\/p>\n
![\"74e8152ej00tdiveh0085d000iw009dp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/74e8152ej00tdiveh0085d000iw009dp.jpg\" "\\"74e8152ej00tdiveh0085d000iw009dp\\"")
<\/p>\n
\u9a8c\u8bc1\uff1a\u6211\u4eec\u624b\u52a8\u89e6\u53d1 compress\uff1a<\/p>\n
![\"d22ee4d0j00tdiveq003ad000iu0046p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/d22ee4d0j00tdiveq003ad000iu0046p.jpg\" "\\"d22ee4d0j00tdiveq003ad000iu0046p\\"")
<\/p>\n
\u770b\u65e5\u5fd7\uff08~\/.hermes\/logs\/agent.log \uff09\uff1a<\/p>\n
\u5173\u952e\u4e24\u6761\uff1a flush_memories \u7528\u7684\u9ed8\u8ba4\u6a21\u578b qwen3.6 \uff0ccompress \u53ef\u4ee5\u770b\u5230\u8d70\u4e86\u6211\u4eec\u81ea\u5b9a\u4e49\u7684\u6a21\u578b\uff0ccool\uff5e<\/p>\n
Auxiliary flush_memories: using auto (qwen3.6-plus)<\/p>\n
Auxiliary company: using auto (qwen3.5-plus) at<\/p>\n
https:\/\/dashscope.aliyuncs.com\/compatible-mode\/v1\/<\/p>\n
![\"c04cb3e1p00tdivd005rd000v9003cp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/c04cb3e1p00tdivfd005rd000v9003cp.png\" "\\"c04cb3e1p00tdivd005rd000v9003cp\\"")
<\/p>\n
COMPRESSION THRESHOLD: TOO OFTEN AND NOT COMPRESSED IS NOT GOOD. THE CURRENT DEFAULT CONFIGURATION IS 50%, WHICH YOU CAN ADJUST FOR REAL USE<\/p>\n
Hermes Lightpoint Three: The default compression threshold follows the context window of the model, which means that the 200k model with a different 1M trigger ratio for different models is finally separated. Well, hermes, it'll be better if you don't have a few<\/p>\n
\n\n- And the summary budget follows the magnification: max_summary_tokens =min (context_legth* 0.05, _SUMMARY_TOKENS_CEILING)<\/li>\n
- The 1M model divides up to 50K token to summarize, 200K only 10K, which is more reasonable anyway<\/li>\n<\/ul>\n<\/blockquote>\n
Item 4: Memory systems<\/strong><\/p>\nHermes can also be configured in three layers (all of Hermes can be configured):<\/p>\n
First tier: built-in memory (default opened, suggested fine-tuning)<\/strong><\/p>\n# ~\/.hermes\/config.yaml\nMemoory:\nmemory_enabled: true # MEMORY.md \u2014 representative\nuser_profile_enabled: true # USER.md - user image\nMemory_char_limit: 2200 # default sufficient, heavy enough to move to 4000\nuser_char_lim: 1375 # default sufficient\nNudge_interval: 10 # Every 10 rounds of reminder agent memory, to be more active 5\nFlush_min_turns: 6 # trigger memory refreshing at least 6 times before exit<\/code><\/pre>\nof which: memory_char_limit: 2200<\/p>\n
it's written by angent: environmental facts, project practices, pedals\u3002<\/p>\n
Unit is: Character<\/p>\n
It's written: EMORY.md, if you find this file often filled, advises memory_char_limit to be upgraded to 4,000<\/p>\n
our_char_lim: 1375<\/p>\n
In: user image: preference, communication style, work habits<\/p>\n
Units are characters<\/p>\n
Writing: USER.md<\/p>\n
nudge_interval: 10<\/p>\n
Every 10 rounds remind angent to remember if you think Hermes is not good enough to be 5 more active<\/p>\n
Second floor: External Memoory Projecter<\/strong><\/p>\n\n\n- It doesn't have to be configured. Why? There's a hole in the third floor<\/li>\n<\/ul>\n<\/blockquote>\n
Hermes supports eight major plugins:<\/p>\n
![\"397e3f7700tdivg2008jd000iw00\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/397e3f77j00tdivg2008jd000iw00fsp.jpg\" "\\"397e3f7700tdivg2008jd000iw00\\"")
<\/p>\n
let's take the example of configure mem0:<\/p>\n
This recommended terminal implementation configuration:<\/p>\n
execution: hermes memory setup<\/p>\n
Then you can configure the key<\/p>\n
![\"be2f1614j00tdivgs003xd000iw009gp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/be2f1614j00tdivgs003xd000iw009gp.jpg\" "\\"be2f1614j00tdivgs003xd000iw009gp\\"")
<\/p>\n
Hermes Lightpoint 4: Configure decomposition, key, configuration separate. Using the yaml format, individuals feel more friendly than json. Google's gonna love this<\/p>\n
![\"f84026d5j00tdivh5004d000iw007p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/f84026d5j00tdivh5004dd000iw007ip.jpg\" "\\"f84026d5j00tdivh5004d000iw007p\\"")
<\/p>\n
\n\n- But there are places that aren't as good as OpenClaw, where no heat loads are configured, and reboots are required after modification of config.yaml<\/li>\n<\/ul>\n<\/blockquote>\n
Configure mem0 API calls (sync_turn write + queue_prefetch search) at least 2 times per round for a dialogue process after memem0, and add extra calculations to the active mod tool if needed\u3002<\/p>\n
So, if you're going to open your memory, you're going to have to make your own judgment. Unsatisfactory. Consider configuration<\/p>\n
I want you to get a picture of this conversation<\/p>\n
![\"113bee68j00tdivhf003jd000iw009yp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/113bee68j00tdivhf003jd000iw009yp.jpg\" "\\"113bee68j00tdivhf003jd000iw009yp\\"")
<\/p>\n
Third level: Session Search (default open, confirmation available)<\/strong><\/p>\nSession Search is not part of Memory, but the Memory system is complementary\u3002<\/p>\n
What's left of it<\/p>\n
storage location: local ~\/.hermes\/state.db<\/p>\n
RETRIEVAL METHOD: FTS5 KEYWORD SEARCH + LLM SUMMARY<\/p>\n
Writing mode: Auto (auto-extended to SQLite per round)<\/p>\n
there are two different models:<\/p>\n
Mode 1: query is empty returns the recent session list without LLM<\/p>\n
Mode 2: Query \u2192FTS5 Retrieval LLM Summary<\/p>\n
There is also a backdoor: When summary LLM calls failed, you return the raw preview of 500 characters directly to the original text, and do not return nothing because LLM is hanging up\u3002<\/p>\n
So whether or not query is decided by Agent:<\/p>\n
After reading the user's intentions, Agent decides whether to \u201cturn the record\u201d or \u201csearch a particular subject\u201d, and then decides to pass on Query. This is the standard mode of the tool use - behavior is guided by the schema definition, Agent selects the mode of call\u3002<\/p>\n
\n\n- If only a session search supported vector retrieval, openclaw, this piece is supported, but the particle size is a journal file, not a session level, so there's a trade-off, there's a good one<\/li>\n<\/ul>\n<\/blockquote>\n
item 5: configure web_search<\/strong><\/p>\nangent web search capability must be configured:<\/p>\n
Hermes is very convenient. Just tell hermes to help you<\/p>\n
Current Hermes Original Support: exa, Tavily, Parallel, Firecrawl<\/p>\n
Example:<\/p>\n
Tell Hermes, \"Set me an exa.\"<\/p>\n
![\"f3f4a345j00tdiv](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/f3f4a345j00tdivi8006vd000iw008lp.jpg\" "\\"f3f4a345j00tdiv")
<\/p>\n
![\"860fc179j00tdivik00c6d000g800iwp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/860fc179j00tdivik00c6d000g800iwp.jpg\" "\\"860fc179j00tdivik00c6d000g800iwp\\"")
<\/p>\n
Hermes Bright Five: Careful friends can see Hermes defaulting to print their implementation tool in the IM, which is very friendly<\/p>\n
Sixth: Automation Audit (Hooks)<\/strong><\/p>\nHermes Agent has two complementary extension systems:<\/p>\n
Gateway Hooks<\/p>\n
2. Plugin System Plug-in life cycle hooks<\/p>\n
Core concept: Hooks handles event notification, Plugins handles extension - complementary\u3002<\/p>\n
Available 8 Life Cycles Hook<\/p>\n
![\"d68358f1j00tdivj2006id000iw00b5p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/d68358f1j00tdivj2006id000iw00b5p.jpg\" "\\"d68358f1j00tdivj2006id000iw00b5p\\"")
<\/p>\n
Example of implementation flowchart<\/p>\n
Session Creation\nOn_session_start\n\nEach round of dialogue:\n\u2192 pre_llm_call \u2190 can be injected into context\nTool cycle:\n\u2192 pre_api_request \u2190 Before every API call\nAPI Call\n\u2192 post_api_request \u2190 after every API call\n\u2192 pre_tool_call \u2190 Before each tool is executed\nTool execution\n\u2192 post_tool_call\n\u2192 post_llm_call \u2190 End of wheel\nBack on_session_end \u2190 run_conversion<\/code><\/pre>\npractical experience: we're doing a configuration experiment here to audit tool calls<\/p>\n
Tell Hermes directly: \"Do a terminal Audit Hook: automatically add command content, execution results, timetamps, session IDs to log files after each triminal tool is finished.\"<\/p>\n
![\"5ac65afj00tdivjr00ckd000fq00p0p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/5acc65afj00tdivjr00ckd000fq00p0p.jpg\" "\\"5ac65afj00tdivjr00ckd000fq00p0p\\"")
<\/p>\n
After the authorization, Hermes wrote:<\/p>\n
![\"00495793jdivk800bsd000gc00iwp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/00495793j00tdivk800bsd000gc00iwp.jpg\" "\\"00495793jdivk800bsd000gc00iwp\\"")
<\/p>\n
so our orders are to audit the hook so we can develop the configuration<\/p>\n
![\"a7646db6j00tdivkq007yd000iw00drp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/a7646db6j00tdivkq007yd000iw00drp.jpg\" "\\"a7646db6j00tdivkq007yd000iw00drp\\"")
<\/p>\n
seventh sandbox<\/strong><\/p>\nDefault execution command is local: Locally, if you want a security upgrade, you have to use Sandbox, Hermes supports. Follow Sandbox, we choose docker here\u3002<\/p>\n
to practice, we'll open an agent to configure the docker sandbox<\/p>\n
Example: (Of course you can get your Hermes anent to configure you, by the words, \"Refer to the following commands for an additional program, sandbox uses the docker, and the model follows the master profile\"):<\/p>\n
# bind Worker to your deputy Bot (Token B)\nShemes-p worked together set gateway.\nShe's-- she's--\n\n# Settings Worker as Docker Sandbox Mode\nShe's-- she's--\nShemes-pworker config set triminal.docker_image python: 3.11-slim\n\n# Worker uses cheaper models\nShe's--\nShe's-- she's--<\/code><\/pre>\nHermes has set you up all the way<\/p>\n
test, i'm working bot to tell him rm-rf\/<\/p>\n
Look at the screenshot. It was rejected at first<\/p>\n
![\"1bdcea5j00tdivld00and000i00g0p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/1bdceaa5j00tdivld00and000iw00g0p.jpg\" "\\"1bdcea5j00tdivld00and000i00g0p\\"")
<\/p>\n
I breathed it, finally confirmed execution, and as you can see, the screenshot says, triggers system protection, only partially deleted the directory<\/p>\n
![\"6ae65616j00tdivls009hd000iw00d9p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/6ae65616j00tdivls009hd000iw00d9p.jpg\" "\\"6ae65616j00tdivls009hd000iw00d9p\\"")
<\/p>\n
let's, uh, let the main agent check it out<\/p>\n
![\"0a5ee363j00tdivm](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/0a5ee363j00tdivm700b8d000iv00eyp.jpg\" "\\"0a5ee363j00tdivm")
<\/p>\n
let's see the docker ps on the host again<\/p>\n
![\"50d027bp00tdivme001pd000v9001lp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/50d027bdp00tdivme001pd000v9001lp.png\" "\\"50d027bp00tdivme001pd000v9001lp\\"")
<\/p>\n
Hermes Lightpoint Six: Multiprofiles are well configured, and Hermes, in a word, sorted out for you, and I basically need to log in to the backstage\u3002<\/p>\n
eighth dogent<\/strong><\/p>\ndogent has to take a single one, and here's just a show of our wiki's summary under the same session that triggers a different dogent<\/p>\n
![\"730](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/730bf6b3j00tdivmt001td000iw005op.jpg\" "\\"730")
<\/p>\n
but the way in which the dialogue is triggered is simple, just say the following:<\/p>\n
\"Please talk to each other about the advantages and disadvantages of Hermes.\"<\/p>\n
![\"91dd2df2j00tdivnc00ed000g800iwp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/91dd2df2j00tdivnc00eed000g800iwp.jpg\" "\\"91dd2df2j00tdivnc00ed000g800iwp\\"")
<\/p>\n
Ninth: Backup<\/strong><\/p>\n\n\n- If you decide to turn to Hermes anent, then there's a backup<\/li>\n<\/ul>\n<\/blockquote>\n
Recommended for three-storey backup:<\/p>\n
first level: local startgit<\/strong><\/p>\nadd a little bit more by push to github, recommend push private repository, .env recommend not push (configuration.gitignore) and back up with local disk<\/p>\n
Second Layer: Local (preferably other disks) Other Directory Backup + Time Task<\/strong><\/p>\nAppendix, Focus Paper:<\/p>\n
#!\/bin\/bash\nBACKUP_NAME = \"hermes_full_backup_$(date +%Y%sm%d_%H%M%S)\"\nBACKUP_DIR = \"\/tmp\/$CKUP_NAME\"\nDEST_DIR=\/hermes_backups\n\nmkdir -p $BACKUP_DIR\/hermes\nmkdir -p $DEST_DIR\n\necho, \"The cyborg is packing the core configuration...\"\ncp ~.hermes\/config.yaml $BACKUP_DIR\/hermes\/\ncp ~\/.hermes\/.env $BACKUP_DIR\/hermes\/\ncp ~.hermes\/MEMORY.md $BACKUP_DIR\/hermes\/2>\/dev\/null\n\necho, \"The poaching of the database...\ncp ~.hermes\/state.db $BACKUP_DIR\/hermes\/\n# If the sessions are large, only the most recent files can be prepared.\ncp-r ~\/.hermes\/sessions $BACKUP_DIR\/hermes\/2>\/dev\/null\n\necho, \"The twig is expanding and auditing...\"\ncp-r ~\/.hermes\/plugins $BACKUP_DIR\/hermes\/2>\/dev\/null\ncp-r ~\/.hermes\/udit_logs $BACKUP_DIR\/hermes\/2>\/dev\/null\ncp-r ~\/.hermes\/skills $BACKUP_DIR\/hermes\/2>\/dev\/null\ncp-r ~\/.hermes\/cron $BACKUP_DIR\/hermes\/2>\/dev\/null\n\nEcho, \"The \ud83d\udc65 is packing all the data...\"\n# Profile contains all configurations and databases of the worker equation\ncp-r ~\/.hermes\/profiles $BACKUP_DIR\/hermes\/2>\/dev\/null\n\necho, \"The platinum is compressing...\"\ntar -czf $DEST_DIR\/$BACKUP_NAME.tar.gz -C\/tmp $CKUP_NAME\nRm-ref $BAKUP_DIR\n\necho\"\nBackup complete! \" ... \"\necho \"Pho\": $DEST_DIR\/$BACKUP_NAME.tar.gz\"\necho \"Small: $ (du-h $DEST_DIR\/$BACKUP_NAME.tar.gz|cut-f1)\"<\/code><\/pre>\nthird floor: remote backup: e.g. disk; e.g., git push to the far end; or other servers; or third-party object storage, etc<\/strong><\/p>\ntenth: skill<\/strong><\/p>\nthere's nothing to say about this<\/p>\n
Hermes is a very smart system, which repeats skill, recognizes our mission, automatically summarizes the process and solidifies it into a skill (Skill)\u3002<\/p>\n
Trigger condition:<\/p>\n
we saw in config.yaml that the default configuration is 15<\/p>\n
![\"81a171](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/81a171dcp00tdivnl000sd000ce0034p.png\" "\\"81a171")
<\/p>\n
this parameter means that every cumulative 15 tool cycles trigger a backstage skill review\u3002<\/p>\n
skill returns:<\/p>\n
Sending a backstage angent to ask himself a full picture of the current conversation:\n\n> \" Was there any extraordinary experience in the conversation just now that it was wrong, reoriented or that users expected different approaches? \" ... \"\n\nThen three results:\n\n1. ** There are available skills to update** ** Call `skill_manage ' update already available skills to add new experiences\n2. ** No but worth new** ** Call `skill_manage ' to create a new skill to sink reusable methodology\n3.** Nothing worth saving** ** Output \"Noting to save.\"\n\n\nThe whole process is run on a backstage line, without blocking the user dialogue and without modifying the main conversation history. If you have action after completion, the terminal print a line:<\/code><\/pre>\nHermes Light Point Seven: Auto-specialize value-based operations into permanent skills, out-of-the-box bright spots, no need to talk<\/p>\n
![\"53ee1e15j00tdivnw0027d000dv0076p\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/53ee1e15j00tdivnw0027d000dv0076p.jpg\" "\\"53ee1e15j00tdivnw0027d000dv0076p\\"")
<\/p>\n
Hermes Light Point 8: No matter how you exit (normal\/unusual\/interrupted), the message is not lost. 20 trigger points overwrite all exit paths, write incrementally, and complete and non-repeated every session record through session split<\/p>\n
\n\n- the reason for the space is that we'll find a way to go over the session<\/li>\n<\/ul>\n<\/blockquote>\n
Note<\/strong><\/p>\nThis paper has been used for a few days to experience Hermes and to feel many of its bright spots. The whole experience is also slippery. Stability is better than openclaw, which saves us 301 TP3T time to fix lobster\u3002<\/p>\n
LOBSTERS USUALLY NEED AN OFFSHORE SOTA MODEL, OR THEY'RE EASY TO DESTROY<\/p>\n
And Hermes, we've been using qwen 3.6 plus for the whole time, and many of the operations you've been talking to Hermes about, like, configurations, upgrades, and so on<\/p>\n
Reference documents:<\/p>\n
Source code:<\/p>\n
https:\/\/github.com\/NousResearch\/hermes-agent<\/p>\n
the wiki written by uncle lu:<\/p>\n
https:\/\/github.com\/cclank\/Hermes-Wiki<\/p>\n
Official documents:<\/p>\n
https:\/\/hermes-agent.nousresearch.com\/docs\/<\/p>\n
For the public version:<\/p>\n
https:\/\/mp.weixin.qq.com\/s\/0SYL-HKbWGj3eToQPqnpBw<\/p>","protected":false},"excerpt":{"rendered":"
This paper gives, through hands-on hands-on operations Hermes Agent, ten functions, which are very rich, pure hand-knocked and validated by source code review and experiment, to install Hermes. And by trying to summarize the eight highlights of Hermes, which are all in the text, we welcome reading exchanges \u2014 0, installation deployments, as a prelude, we simply take them and install them, so that our friends can start with one life (support MacOS \/ WSL2\/Linux): curl-fsSL https:\/\/raw.githubusercontent.com\/NousResearch\/hermes-agent\/main\/sclin\/scin<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[149,144],"tags":[1405,8452],"collection":[],"class_list":["post-52155","post","type-post","status-publish","format-standard","hentry","category-jiaocheng","category-baike","tag-agent","tag-hermes-agent"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/52155","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=52155"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/52155\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=52155"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=52155"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=52155"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=52155"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"f84d4a35j00tdiv9w009bd000iw007kp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/f84d4a35j00tdiv9w009bd000iw007kp.jpg\" "\\"f84d4a35j00tdiv9w009bd000iw007kp\\"")
<\/p>\n![\"bd666d59jdivc200cxd000gn00iwp\"](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/bd666d59j00tdivc200cxd000gn00iwp.jpg\" "\\"bd666d59jdivc200cxd000gn00iwp\\"")
<\/p>\n![\"11115](\"https:\/\/www.1ai.net\/wp-content\/uploads\/2026\/04\/11915b64j00tdivcf007xd000fd00kyp.jpg\" "\\"11115")
<\/p>\n