I am launching Relaticle here next Tuesday. It is an open-source CRM where AI agents can do real work on your data, and I want feedback on one design decision before launch day.
Relaticle has two trust levels. External agents (Claude Desktop, Cursor, any MCP client) authenticate over OAuth and write through 37 tools directly, like an API client. The in-app assistant is stricter: every write it wants to make becomes a proposal card, and a person approves or skips it record by record before anything lands.
The question I keep going back and forth on: is per-record approval the right granularity, or does it become noise once you trust the assistant? Would you want a "trust this kind of change" switch, or does that defeat the point?
If you run a CRM with any AI in it today, I would like to hear where you draw that line. The whole thing is AGPL and self-hostable, including the model via Ollama, so you can look at how it works rather than take my word for it.
Hi Product Hunt, Manuk here. I've been building Relaticle for about two years, mostly solo.
It started as a conventional open-source CRM. Then I added agents and hit a question I couldn't dodge: how much authority should each kind of client get? Read-only access was too limited to be useful. Silent writes into customer data felt wrong.
I ended up with two trust levels. External MCP clients (Claude, Cursor, anything that speaks MCP) authenticate over OAuth and write directly, the same way an API client would. The in-app assistant works next to human users, so every write it wants becomes a proposal that shows the exact change and waits for approval. Batches are reviewed record by record: you approve or skip each one, and whatever you approved is kept.
The chat client itself was rebuilt after failures I hit in real use. A rate-limited retry could delete a conversation turn. A stale stream handler could write a draft into the wrong conversation. Drafts now survive reloads, failed sends stay visible, and answers link back to the records they mention.
Everything self-hosts under AGPL, including model inference with Ollama. Each workspace defines its own custom fields, and they show up in the agents' schemas automatically.
One question I'm still chewing on and would genuinely like opinions on: should an in-app assistant ever get policy-based auto-approval, or should every write stay explicit?
@manukminasyan as someone who deals with messy client data all day an AI that acts as a strict proposal engine is exactly what market need. congrats for launch🙌
@priya_kushwaha1 Thanks, Priya! Glad this resonates with someone working with client data every day.
@manukminasyan For sure.. keep up the great work with it
@manukminasyan Congratulations on the Product Hunt launch! Wishing Relaticle a fantastic launch and lots of success ahead.
@charan_t_m Thanks, Charan! Really appreciate the kind words.
@manukminasyan the two differents trust level for ai actions make a plenty of sense, that could make the setup much easier for team with their own CRM architecture.
what was the challenging chat issue you had to fix during the restructure?
@manukminasyan Approval - gated AI sounds great but who approves the approver ?
Can you set rules so certain fields can be automatically written while others always require approval?
@naimz Not currently, Naim. The assistant can update your fields, but those changes still need approval. What workflow did you have in mind?
@naimz @solozero I like the idea of earning autonomy over time. I’d want people to choose what it can update on its own. Which updates would you trust it with first?
Honestly, this is the kind of CRM + AI approach that makes sense to me.
I like that the AI can actually do things instead of just generating text, but you still get to approve what changes before it touches your data.
And making the whole thing open-source + self-hostable is a huge plus. Nice work!!
@ilya_pashayan Thanks, Ilya! I wanted the assistant to do useful work while keeping the person using it in control. Glad that came through.
This project has a solid based of Laravel that have helped me learn Laravel. I learned so much about how to setup Laravel in a professional - enterprise grade setup which alone is gold. In combination with AI - MCP, it really can turn a basic CRM into a full-fledge internal processing system, ie: an ERP.
Wish alot of success for Manuk for the launch!
@laurnts Thank you, Laurentius! Hearing that the codebase helped you learn Laravel means a lot to me. Thanks for the launch support!
Would love an undo trail for approved writes, not just the approve step. Excited to see where this goes!
@colegawin Thanks, Cole! Being able to undo a change after approving it makes sense. Approval doesn’t mean you’ll never need to take it back. Appreciate the suggestion!
@manukminasyan Manuk jan, congrats on the launch!
It’s great to see a CRM that isn’t just adding AI on top, but is actually designed around AI from the ground up. And the MCP compatibility is really cool stuff!
Good luck =)
@adana Thanks, Adana jan! Glad the AI approach and MCP stood out to you. Really appreciate you being here for the launch!
I am also looking for open source crm. Love to explore more.
Congratulations the launch btw