Hermes Agent now has Tool Search, so your agent only loads what it needs

As always, 'hermes update' More info:

Many are wanting this sir 👀👀

👨‍🏫

Very Cool, you guys are adding so many features I can't keep up, and Ive added support for Hermes Agents into My Virtual Office. Love seeing my Hermes agents work together like video game characters 😅💯

Yall won’t stop cooking instant hermes update on my CLI

See this is the difference between Hermes and OpenClaw. Your updates are fewer, more intentional and much more beneficial to users. Plus, they don't break the harness *every* *time*, even with modified source code.

We are not tired of winning with Hermez

This is the right direction for agents. Tool search is powerful, but in production you still need visibility into what got loaded, what got called, and what each step cost.

Use ChatGPT to help. 🚨 AI scams just leveled up — and they’re coming for everyone. 🎭 Voice Clones 🎥 Deepfakes 🪪 Fake Identities 💸 Scams at Scale The old defenses are dead. 🪽 Nous’ Angels is fighting back. ⚡ AI vs AI Defense 🔍 Evidence-Driven Investigations 🔐 Wallet-Gated Access 👤 Human Verified Outcomes This isn’t just protection. It’s the next line of defense. $ANGELS CA: F2iqrq41NfJoZnK68UBmrbZrLPkPDaJzZHyZFbDdpump

It’s only been a day since I asked for it in reply to v0.15 announcement, and though I know it’s been brewing for a while and it’s not just me, the PERCEIVED speed of response is very impressive. Kudos for listening to your users!

Smart move to reduce context bloat from loading unnecessary tools. Thank you

Context windows or token consumption stopped being a problem with Hermes, overnight heavy work using only 74 million tokens with 173 million cached.

waiting for ya'll to add tool simulations, too ;)

Dumping 50 unused tool schemas into the system prompt is a crime against the KV cache. Dynamic loading is basically the only way agents scale without killing your latency budget.

At this point, I price in that Nous and community will add in an additional solid layer on top of its already solid foundation. Hermes Agent just keeps improving

BWAHAAHHAA I need this. I would like have 250+ MCPs same time

non stop evolution

Hermes has changed my life!!!

oh interesting, this turns tool loading into a retrieval problem. the risk is recall, if the right tool ranks below the cutoff the agent can't call it and won't know it existed. search matches on how each tool is described, so a vague description makes one basically invisible.

nice. Thank you

ok this is huge! Biggest announcement for ai in a long time

Hermes Agent just got a game-changing upgrade: Tool Search! 🔥 No more bloating your context with every tool. Now it only loads what it actually needs. Before: 16 tools = 41% context used After: Same 16 tools = just 3% context On-demand loading done right. Massive token savings. Real scalability. The future of agents is here.

Made by Hermes itself

@brooksy4503 it will be great if it can automatically unload tools,it will be more token efficient

“Exactly that, but with auditing — that’s what Glyph Protocol does.”

I’ve been spamming hermes update more than chatting with hermes recently.

We added the same thing in Ovalpod like 2 months back by training a specialised model just on the Tau bench and using it to filter out tools

interested on how this is implemented. does someone have a link to the pr?

does the trace show which tools were considered but not loaded? once tools become searchable, the skipped choices matter for audit too.

do you guys ever sleep

Can you make telegram reactions visible to the agent?

this is the context management decision I've been waiting for. MCP schemas eating your context window turn by turn — fixed. BM25 retrieval + activation gate handles it silently. the part worth noting: core tools never defer. terminal, memory, file ops — always loaded. everything else — searched on demand. clean architecture. hermes update

cool I’ll have that implemented and make it even faster

Noice

Finally