MakersClaw
Hire AI employees that live in your Slack, Teams, Telegram
Start new thread

MakersClaw - Hire AI employees that live in your Slack, Teams, Telegram

by•
Hire AI employees that run 24/7 in their own container with their own memory. One-click into your Slack, Telegram, or Teams. Pre-built for support, sales, research, SEO, or anything you write yourself. Pay per call for the tools they use.

Add a comment

Replies

Best
Maker
📌
Hey PH 👋 Shreyans here, co-founder of MakersClaw. Sachin's in the thread with me today. We started this because every "hire an AI agent" tool we tried felt like a chat widget with a coat of paint. It forgot the conversation when you closed the tab, it couldn't actually do anything in the apps you use all day, and we kept hitting walls trying to make one do real work. So we built MakersClaw the way we wanted to use it. You hire an AI employee for whatever role you need: support, sales, personal assistant, research, SEO, or your own custom thing. Each one runs in its own container with its own memory, 24/7. You connect it to your Slack, Telegram, or Teams in one click. No bot tokens, no webhook config, no JSON. Sachin's dropping a comment right below with how the guts work, the tools, the runtimes, the pay-per-call model. Read that if you want the mechanics. Two specific things we'd love your honest take on: 1. Does the per-call tool model make sense to you, or is the mental shift from "subscribe to a tool" to "your agent pays per action" confusing the first time you see it? 2. We ship with four pre-built templates (support, sales, personal assistant, SEO). Which one would you actually try first, and which role do you wish we had a template for? We'll be here all day. Pile on the questions and we'll answer everything.

 congrats on the launch! I'd definitely try the PA and the support template first. On sales, I'd actually be a little wary of having the agent interact directly with prospects. What are the guardrails/quality controls you have around that?

Maker

 Thanks Denitsa! 🙏 Totally fair instinct — PA and support are the easiest day-one wins. On sales, the short answer: it's human-in-the-loop by default (drafts, doesn't send, until you trust it), it only speaks from the context you give it so it's not improvising claims, and you set hard limits on what it can't do plus when it hands off to a human. Every message is logged too, so it's never a black box. The goal isn't to replace your judgment on prospects — just to kill the busywork around it. Happy to walk you through the controls if you want to dig in.

Maker

 Thank you, please give it a try and let me know how was the experience :)

 Congrats, team! The one-click connect is the winning flow here. Wrt the templates: SEO piqued my interest instantly, especially if it includes AEO/GEO.

Maker
we have added capabilities and skills to the agent which gets it suited to any task. We tried for AEO/GEO, thanks to the skills understands the task really well.
Hunter

 Many congratulations Shreyans and Sachin on the launch! Happy to see you back on PH :)

The AI employee idea is useful especially with the one-click setup on Slack, Teams and Telegram. I also like that it has memory and pay-per-use tooling, which makes it feel more practical than most bots.

Maker

Hey PH 👋 Sachin here, jumping in with the engineering side for anyone curious.

Each employee is a Kubernetes pod with its own filesystem and its own postgres-backed memory. State survives restarts and channel disconnects. Even a full redeploy. We chose this over a serverless function model because we wanted the agent to be a process you can talk to at 3 AM and have it remember the conversation from yesterday morning. Cold starts and stateless containers kill that.

For app integrations we run a hosted MCP layer at the workspace level. You OAuth once per app (GitHub, HubSpot, Zendesk, Jira, Asana, Airtable, Gmail, Outlook, Calendar, more) and any employee in your workspace can use the integration after that. Each MCP server is managed on our side, so there's no JSON config or token paste on yours. No re-auth per agent.

For configuration we built a chat-driven onboarding flow. Instead of filling forms to set up the employee's role, tone, and context, you talk to it. It asks the questions, you answer, it writes its own config record. The mental model is onboarding a remote hire rather than setting up software.

Skills are modular blocks of context the agent retrieves dynamically when a task needs them. Private skills stay scoped to your workspace. Public ones get installable by any maker. So the agent isn't carrying the whole brain on every call. It pulls the right context for the work at hand.

Two runtimes:
- PicoClaw runs on Python. Lighter, supports email channel and cron scheduling.
- Moltis runs on Rust. Heavier. Web dashboard, browser automation, voice (15+ TTS/STT providers), CalDAV.

Happy to go deeper on any of the architecture. Ask me anything.

the 'ai employee' framing is getting crowded fast - so many tools launched this month doing the same slack/teams bot thing. the per-call pricing on tools is the bit that'll catch people off guard when an agent loops or retries unexpectedly. what actually differentiates the memory layer here vs just wiring up a standard agent with a slack connector?

Maker

 Honestly, fair points. We're not trying to win the "another Slack bot" race - Slack/Teams is just the easy on-ramp. The actual bet is the ecosystem around the agents. On memory and the per-call cost problem you flagged, that's exactly what having a few ai researchers on team helps with, testing a few approaches to optimize and cut cost. No grand claims yet; we'd rather ship something that holds up than overstate it on launch day. Would genuinely value your eyes on it as it evolves.

AI employees living right in Slack and Teams sounds really useful. Can each AI employee be customized to a specific role or task?

Maker
yes we have a vibe based agent creation system which would let you customise the agent to your needs 👍

Congrats on the launch!

The "AI employee in Slack/Teams" framing has been tried a few times but the pricing always trips it up — per-seat feels wrong for a non-human. How did you land on your model?


(Indie maker here, curious about the pricing call more than the product itself.)

Maker

 Thanks! Per-seat felt wrong to us too. We went per-employee instead after trying out multiple pricing models as the product was developing. One monthly subscription per AI worker for its pod and storage. Inference is metered separately against a wallet at provider cost, no markup. Predictable infra floor, and a variable thinking cost you control.

Honestly still chewing on whether one wallet across a customer's whole workforce beats one wallet per employee.

 Per-employee wallets I'd avoid — every business buyer I know hates managing N separate balances. The mental tax outweighs any control they think they're buying.

One shared wallet + admin-set per-employee daily/monthly caps is the model that keeps surfacing in B2B usage-based pricing (Anthropic's API workbench does something like this, AWS budgets too).
You get predictability of "we won't exceed X this month" without the admin overhead of N wallets.


Lift on shared-wallet + caps is also lower than it sounds — one balance, N constraint checks at consume time.

Best of luck with the launch 🙌

Maker

 Human facing is still one subscription and one wallet shared across AI employees 🙌

   Feels like shared wallet + per-employee caps tends to work well in B2B - simpler for finance (one line item, guardrails vs. N balances). Is that what you’re seeing with customers so far?

Giving each employee its own pod with persistent postgres-backed memory that survives restarts is the detail that won me over, "remembers yesterday's conversation at 3 AM" is exactly where most chat-widget agents fall apart. On the pay-per-call model: do users get spend caps or alerts per employee, so a retry loop on a tool can't quietly rack up cost overnight?

Maker

 Tool calls are fail safe, you pay for the successful ones. We are designing makersclaw to be human centric so humans can monitor everything agent does.

AI employees living where the team already talks is the right move. The hard part is making them helpful without becoming one more coworker to manage.

Maker
ohhh I would be really interested in what you think would make ai employee easier 🙏
Maker

 We lean on persistent memory and the file system to cut the reexplaining context cost. You tell it once (or upload context files once), then the agent remembers.

None of this solves the deeper concern fully. The honest version is that "an AI employee that needs zero oversight" isn't a 2026 reality for anyone. We're trying to push the oversight cost as low as possible per action, but it's still genuinely a hard problem.

Super cool idea. I think the pay-per-call tool model is great.

Maker

@adam_maceachern1 thank you, I think it makes it easier to keep adding more while keeping an eye on real spends

The per-call model is a smart call, pay for what your agent actually does is way easier to justify than another monthly subscription. Personal assistant template first for me. Would love to see a research analyst role next

Maker
easier to understand and see value for. Noted for the personal assistant template 🙏
Maker

 Research analyst is the one we've been talking about for the next template drop. We held off in the first batch because the role means wildly different things across industries (financial, market, technical etc.). The first four templates were the ones with clearest scope. The next batch will be shaped by what people actually ask for, so this is genuinely useful signal.

The "configure by chatting with the employee" concept really caught my attention. In practice, do users prefer conversational setup over traditional forms and settings?

Maker

 Yes, from what we're seeing. The conversational setup is the part people are most excited about today, which feels promising. Small dataset though, and the signal could pivot either way as usage scales.

Worth noting: we're already hybrid today. All of the configuration happens through chat, but some settings (the structured ones like model selection, skill selection) are also selectable. So users get to pick the path that fits the task. We might keep it that way long-term rather than forcing everything into chat.

 The hybrid approach sounds like the best of both worlds. Chat for flexibility and settings for precision. Curious to see how this evolves as more users come in!

123
Next