Agent CRM
Customer support management simplified
Start new thread

Agent CRM - Customer support management simplified

by•
Support teams waste time managing emails instead of solving problems. Replies get lost, response times suffer. AgentCRM consolidates everything into one place. Emails become organized cases. Reply directly from the app instead of switching tools. AI suggests solutions to speed up resolution. Automatic translations for global customers. Built for teams of any size—from 2 agents to unlimited. Free to start, pay as you grow. Currently in beta. Join now to help shape the product.

Add a comment

Replies

Best
Maker
📌
Hey everyone! I'm Fares, and I built AgentCRM because I spent years managing customer support teams and saw the same problem everywhere: teams drowning in emails, replies getting lost, and agents burning out on repetitive work. I managed customer success at Cannvisa and saw firsthand how scattered communication killed productivity. Emails were everywhere—Gmail, Outlook, spreadsheets. Customers waited days for responses because replies got buried. So I built the tool I needed: a single dashboard where emails automatically become cases, your team replies without switching apps, and AI actually suggests solutions instead of just summarizing problems. The journey was messy. Started on Railway, ran into memory issues, moved to Vercel. Spent weeks on security (sessionStorage, encryption, proper OAuth). But now it works reliably and the feedback from early testers has been great. I'm targeting $1-2K MRR in the next 6 months and actively building based on what teams ask for. Canned responses and call features are next. Looking forward to your feedback. What's broken? What's missing? Let me know.

pasting a ticket into Claude already gives me a solid draft so unless this "agent" executes state changes directly in the databse instead of just suggesting text it is just a UI layer over a prompt, what actual account actions does it resolve end to end or

Maker

 
Fair point, and I'd rather be straight about it: today, the AI suggests — it doesn't execute. It drafts replies, summarizes threads, and translates, with the customer's full ticket history already loaded in, so you're not copy-pasting context into a separate tool every time. That's real time saved, but you're right that it's not agentic in the "takes actions on your behalf" sense yet.

What's actually next on the roadmap is closing exactly that gap — having the agent execute directly against the database instead of just producing text: auto-updating ticket status (open → pending → resolved) based on conversation signals, auto-tagging and auto-routing tickets to the right agent, auto-creating follow-up tasks, and auto-sending replies for a defined set of FAQ-type cases (behind a human-approval toggle at first). That's the difference between a drafting tool and an actual agent, and it's the next milestone after this beta feedback round.

Appreciate you pushing on this — it's the right question to ask before "AI Agent" becomes a marketing word instead of a feature.

 preloading the ticket context saves some copying. the real "leverage" only starts once it can safely mutate the backend state directly. want to test the execution layer

Maker

 
Here's what I'm building toward: autonomous ticket-status transitions. The idea is that when a customer replies to an existing ticket for example ("It is fixed now. Thanks!"), the system reads the message and, if it clearly signals the issue is resolved, moves the ticket to Resolved directly — no one has to open it and click a button.

That's the gap I'm targeting specifically. Tools like HubSpot or Intercom will draft you a reply or summarize a thread, but the resolve/tag/route/follow-up-task decision still sits entirely on the agent — the software hands you information and waits. The bet is that a meaningful chunk of that decision-making is mechanical enough to hand off, with a human only stepping in for the ambiguous cases. I've got the logic built and I'm testing it against real replies right now — not calling it done until I've seen it actually behave correctly, not just deployed. Since you're the one pushing on whether this is real, want to be an early tester once it's confirmed solid? I'd rather have someone trying to break it than assume it holds up.

 Let me know when autonomous transitions reach beta. I want to test how you handle false positives and rollback safety on auto-resolves.