Sudhir Mantena

2.2K posts

Sudhir Mantena banner
Sudhir Mantena

Sudhir Mantena

@smantena

Goal: Help 1 million Professionals & Businesses use Personal AI Agents Built Products @Paytm, Simpl, @JioHotstar, @Cleartrip, OwnStartup, @Bloomberg & @Intel.

Bengaluru, India Katılım Ağustos 2008
1.2K Takip Edilen427 Takipçiler
Sudhir Mantena
Sudhir Mantena@smantena·
@Saboo_Shubham_ Curious, how do you solve multi-agent memory? Gbrain or something off-the-shelf or custom built?
English
1
0
0
33
Shubham Saboo
Shubham Saboo@Saboo_Shubham_·
Codex /goal builds it. Claude Code /goal review and refines it. Hermes /goal manages the orchestration and handoff. All tracked on a single Kanban Board and agents keep running in the loop.
English
54
38
802
82.1K
Sudhir Mantena
Sudhir Mantena@smantena·
Hi @Voxyz_ai Very thoughtful writeup, thanks for sharing. I’m still early in learning how memory should be designed for AI systems, but I’m thinking about it in layers: 1. Identity: who 2. Episodic: what happened 3. Semantic: facts and knowledge 4. Procedural: how it was done 5. Organizational: docs, tickets, discussions, etc. How does your setup handle these layers?
English
0
0
0
23
Sudhir Mantena
Sudhir Mantena@smantena·
@madhujdhav23 I opened the website. There's barely any info on what it does. Scary for people to scan a QR code and open whatsapp given all the scams that are doing on.
English
0
0
0
16
Sudhir Mantena
Sudhir Mantena@smantena·
Hermes ⇔ Whatsapp: How do you get the Agent to STOP talking? Your Hermes agent ⇔ Whatsapp bot is now working. When you DM the bot on Whatsapp, it works great. Now you add the bot to a Whatsapp group. By default, it wouldn't stop talking. It will respond to every message, even those sent to other members in the group. How do you get the bot to stop responding to all messages? In your ~/.hermes/config.yaml Under whatsapp, channel_prompts: require_mention: true Now only those messages tagged with @ bot_num, the bot will respond. #hermesAIAgent
English
1
0
1
65
Sudhir Mantena
Sudhir Mantena@smantena·
@DylSwanepoel @DylSwanepoel I am deeply interested in this topic. Do you know companies that have built such org-wide memory systems and what have they used to build agents & memory systems?
English
1
0
1
16
Dylan Swanepoel
Dylan Swanepoel@DylSwanepoel·
@jmwind Institutional memory is turning from tribal knowledge into infrastructure. The companies that win probably won’t be the ones with the best models. They’ll be the ones with the best memory systems.
English
1
0
1
623
Jean-Michel Lemieux
Joined a new AI-native company this week and it’s kind of wild how different it feels already. The laptop arrived, I logged in, and an agent basically took over from there. It set up my dev env, pulled repos, fixed dependency issues, got permissions approved, pointed me at the backlog, linked the architecture docs, and surfaced the Slack debates I actually needed to read before touching production. When I needed context on something, I asked the agent and it found the exact thread from months ago explaining why a decision was made, who owned it, the related Linear issues, and the PRs connected to it. I’ve only been here 3 days but it honestly feels like I’ve worked here for a year because the usual friction and scavenger hunt for context just isn’t there anymore. We should probably stop calling this “onboarding” and rename it to “mounting” because this feels a lot more like mounting a distributed filesystem called “institutional memory” than slowly getting drip-fed context over 6 months.
English
269
398
6.1K
945.5K
Sudhir Mantena
Sudhir Mantena@smantena·
@jmwind @jmwind Thats' the baseline for new employee onboarding in 2026 with org-wide memory. Am curious, what does your company use to build these Agents + Org-wide memory?
English
0
0
0
267
Sudhir Mantena
Sudhir Mantena@smantena·
I have been using Claude, Gemini & OpenAI all in parallel along with Hermes. Claude has been the best among all to help debug and think through the setup. But even Claude led me astray so many times into rabbit holes wasting tons of time. But these are early days and not using these tools isn’t an option if you want to move fast.
English
0
0
0
146
GREG ISENBERG
GREG ISENBERG@gregisenberg·
If you were the kid who carried every group project in school, you'll love the AI era. People let you down. Agents will too. They break, they hallucinate, they do the wrong thing at 3am. The difference is the feedback loop. When a person drops the ball, you have a conversation, wait a week, and hope it gets better. When an agent drops the ball, you fix the prompt and it's better in 5 minutes. Same frustration with 100x faster resolution. Not everyone will love working like this. But a lot of people will. I really like it.
English
124
26
546
35.4K
Sudhir Mantena
Sudhir Mantena@smantena·
Have been on the $20/mo plan. Have been on GPT 5.4 mini since the start and haven’t hit the limits yet. Want to switch to GPT 5.4 but am worried I will hit the limits quickly. In the experimentation phase will continue with 5.4-mini, but once I have the setup fully working and I know the exact flow, will optimize for Quality vs Tokens.
English
0
0
0
288
Ole Lehmann
Ole Lehmann@itsolelehmann·
The best model setups to run on Hermes (by price tier): 1. If you have infinite budget: Go with GPT 5.5 or Claude Opus 4.7. Both are top class and you'll feel the difference on any non-trivial task. I use GPT 5.5 because of the Codex login. It logs into your actual ChatGPT account so the usage hits your existing subscription (no separate API bill). Anthropic doesn't allow the same for Claude, so if you go with Opus 4.7 you're paying expensive API fees on top of whatever you're already paying for Claude. At OpenAI's $120/mo tier and up, you basically won't hit rate limits during a normal workday. 2. If you have a tighter budget: Run GPT 5.5 with DeepSeek V4 Flash as a fallback for when you blow through your $20/mo ChatGPT limits. If you'd rather never hit limits at all, swap GPT 5.5 for GPT 5.4 mini as your primary model. Or skip ChatGPT entirely and just use DeepSeek V4 Flash on its own and you should come in under $30/mo for a typical month. 3. If you have a beefy local machine: Qwen 3.6 is the best setup. Zero per-token cost once it's running and your code never leaves your machine. Pick the one that matches your monthly spend and start there. Anything I'm missing / other setups you'd recommend?
Ole Lehmann@itsolelehmann

HERMES AGENT FOR DUMMIES Everyone on X keeps talking about Hermes Agent and I finally get why: Once you have an AI that's always-on, remembers everything, and you can just text from your phone, you're never going back to a regular AI chat window. That's Hermes. You text it like an executive assistant and it just handles things. Think of it like an affordable OpenClaw that actually works and is reliable. I was using OpenClaw earlier this year, but it kept breaking and the costs were adding up, so I quit personal agents for a while. Then I set up Hermes and it's what I wanted OpenClaw to be from the start. Here's how it works: > You can talk to it on 19+ messaging platforms (I use Telegram and Discord) > You give it a personality file called SOUL .md so it behaves how you want. > You give access to your email, calendar, full browser access, and whatever other tools you use through integrations. > And you plug in whatever models you want as the brain. I use GPT 5.5 for heavy thinking and DeepSeek for lighter tasks so I'm not burning tokens on every little request (under $20/month in token spend). And I run mine on a Hetzner server for about $5/mo, which enables Hermes to be always-on and persistent. So you can schedule tasks that run autonomously, even while you sleep and even while your devices are off. For example, I have one that scans my email for customer support tickets every morning, matches customer to stripe data, sends out missing links, checks refunds etc. Stuff that used to take me HOURS. And the part that actually separates it from OpenClaw and everything else: Hermes writes its own skills from experience. Every time it completes a task, it saves what worked and turns it into a reusable skill it can run again without you explaining anything. So it literally gets smarter and better over time for your specific workflows. Here's what other people are doing with it: > inbox zero and calendar management from Telegram on their phone > overnight coding that debugs itself while they sleep > a family WhatsApp bot where 5 people share one agent to get stuff done > daily briefings texted every day at 8am > smart home control through Home Assistant > etc Once you try texting an AI that already knows your whole setup and just does things when you ask, opening a chat window and starting from scratch every time feels broken. This is the first AI agent that feels like a legit Chief of Staff.

English
59
24
475
62.3K
Sudhir Mantena
Sudhir Mantena@smantena·
@OpenClawTips @Teknium Time taken to hit the first Aha moment is very short with Hermes compared to OpenClaw. Post setup, Hermes just works without too much maintenance.
English
0
0
0
52
Hermes Agent Tips
Hermes Agent Tips@HermesAgentTips·
People who moved from OpenClaw to Hermes Agent: what was the one thing Hermes fixed that OpenClaw never could?
English
188
8
263
62.6K
Timur Yessenov
Timur Yessenov@Timur_Yessenov·
@smantena @ericosiu @smantena Not through channel_prompts. I’d keep prompts for behavior/context, then use a separate router: channel/group -> agent/profile/policy before the message enters the agent. Custom logic should be a small tool/hook, not hidden inside the prompt.
English
1
0
0
25
ericosiu
ericosiu@ericosiu·
Slack > Telegram/WhatsApp when it comes to using Hermes/OpenClaw for work. A few reasons: Multiplayer - your agents can collaborate with your team. More importantly, your agents can serve as the single brain that connects all your data sources to allow your company to be queryable. When you get your data pulls and strategies 10x faster, you move 10x faster. Not to mention that when your team sees these agents executing on work, it unlocks their brains. Threading - Threaded replies make it easy to then spin out those chats into separate windows to work in parallel. Branching - More often than not, my agents and I will create 'idea babies' which I then want to spawn off into new threads. Message forwarding is basically branching messages off. When you combine the /kanban and /goal commands, things stay organized, and your agents keep working til things get done. What did I miss?
English
6
1
21
2.5K
Sudhir Mantena
Sudhir Mantena@smantena·
@Timur_Yessenov @ericosiu So is there custom python code with routing logic that you trigger from channel_prompts and handle everything inside the code?
English
1
0
1
11
Timur Yessenov
Timur Yessenov@Timur_Yessenov·
@smantena @ericosiu This is exactly the layer I’d separate from prompts: channel/group routing decides which agent profile receives it, then prompts stay smaller. In AnthroClaw I’m leaning toward agent policies + channel routes rather than one giant channel_prompts map.
English
1
0
0
10
Sudhir Mantena
Sudhir Mantena@smantena·
I had the exact same use case. I wanted to create separate profiles for different WhatsApp groups but with a single Hermes instances (multiple means, multiple WhatsApp phone nos). Based on my convo with Hermes lead, you can’t do the above. So instead I customized channel_prompts for each group that I want to behave differently. channel_prompts: default: | behavior details … whatsapp_group_id: | behaviour_details … Let me know if you need more detailed help.
English
1
0
1
44
Timur Yessenov
Timur Yessenov@Timur_Yessenov·
@ericosiu That combo makes sense: Slack for team/shared context, Telegram for quick personal interrupts. Do you keep the same agent identity across both, or split agents by channel/workspace?
English
1
0
0
29
Michael
Michael@Mike_alive·
@itsolelehmann Im interested in the WhatsApp family setup - any best practices there? And why do you use discord AND telegram?
English
4
0
5
2.1K
Ole Lehmann
Ole Lehmann@itsolelehmann·
HERMES AGENT FOR DUMMIES Everyone on X keeps talking about Hermes Agent and I finally get why: Once you have an AI that's always-on, remembers everything, and you can just text from your phone, you're never going back to a regular AI chat window. That's Hermes. You text it like an executive assistant and it just handles things. Think of it like an affordable OpenClaw that actually works and is reliable. I was using OpenClaw earlier this year, but it kept breaking and the costs were adding up, so I quit personal agents for a while. Then I set up Hermes and it's what I wanted OpenClaw to be from the start. Here's how it works: > You can talk to it on 19+ messaging platforms (I use Telegram and Discord) > You give it a personality file called SOUL .md so it behaves how you want. > You give access to your email, calendar, full browser access, and whatever other tools you use through integrations. > And you plug in whatever models you want as the brain. I use GPT 5.5 for heavy thinking and DeepSeek for lighter tasks so I'm not burning tokens on every little request (under $20/month in token spend). And I run mine on a Hetzner server for about $5/mo, which enables Hermes to be always-on and persistent. So you can schedule tasks that run autonomously, even while you sleep and even while your devices are off. For example, I have one that scans my email for customer support tickets every morning, matches customer to stripe data, sends out missing links, checks refunds etc. Stuff that used to take me HOURS. And the part that actually separates it from OpenClaw and everything else: Hermes writes its own skills from experience. Every time it completes a task, it saves what worked and turns it into a reusable skill it can run again without you explaining anything. So it literally gets smarter and better over time for your specific workflows. Here's what other people are doing with it: > inbox zero and calendar management from Telegram on their phone > overnight coding that debugs itself while they sleep > a family WhatsApp bot where 5 people share one agent to get stuff done > daily briefings texted every day at 8am > smart home control through Home Assistant > etc Once you try texting an AI that already knows your whole setup and just does things when you ask, opening a chat window and starting from scratch every time feels broken. This is the first AI agent that feels like a legit Chief of Staff.
English
18
18
306
142.1K
Sudhir Mantena
Sudhir Mantena@smantena·
Create WhatsApp groups for family & work. DMs = ALLOWLIST allowlist: - “17321112222” - “17322223333 unauthorized_dms = ignore WhatsApp_group_policy = ALLOWLIST allowlist: - “groupid-1” - “groupid-2” require_mention = true (most important to make sure it responds only to messages specifically tagged to bot. Else bots starts responding to every msg)
English
0
0
0
72
Sudhir Mantena
Sudhir Mantena@smantena·
@itsolelehmann @Mike_alive Your question about WhatsApp setup: 1/ Single Hermes instance - single WhatsApp no 2/ You can add the Hermes bot in as many WhatsApp groups as you like 3/ By default, bot starts to respond to every msg in the group (which gets annoying quickly) Solution that issue is here x.com/smantena/statu…
English
0
0
0
257
Sudhir Mantena
Sudhir Mantena@smantena·
@Teknium @NousResearch @Teknium Re-clarifying my question clearly: In a single Hermes gateway instance, using a single whatsapp phone number, is the below possible? Whatsappp-group-1 ⇒ Profile-1 Whatsappp-group-2 ⇒ Profile-2
English
1
0
0
67
Sudhir Mantena
Sudhir Mantena@smantena·
@papiofficial Very helpful learnings. I run on Hermes. Recently connected it to Notion instead of Obsidian. I am in the process of launching multi-agents. Let’s see how that goes. Will keep your observations in mind.
English
0
0
1
64