
Embage
Build agent for your customers, run workflows, analyzed.
259 followers
Build agent for your customers, run workflows, analyzed.
259 followers
Build multichannel AI chat and voice agents for your website, Telegram, and beyond. Create specialized sub-agents to orchestrate workflows across your knowledge bases, datastores, and hundreds of third-party tools in real time. Automatically evaluate every interaction, score lead quality, and extract structured data straight from conversation transcripts.












Free Options
Launch Team / Built With

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted



Hi Product Hunt! 👋
I’m Aradhya, maker of Embage.
Full disclosure: I originally built Embage to launch a one-person company. I picked customer support because reps spend way too much time on repetitive tasks, sifting through docs, processing returns, and sending standard emails.
With Embage, AI agents handle all of that autonomously. If anything critical comes up, your human team can step in whenever the agent requests a handoff.
The Vision: My goal with Embage is to give lean teams from e-commerce stores to local service providers, the power to run fully automated, multi-step operations without needing dedicated dev support.
To make this happen beyond basic support, we built Custom Datastores. Datastores let you create custom tables to store whatever data your agents collect, like tickets, leads, or feedback. Simply create a datastore, build your main agent, link a sub-agent with rules on when to use that data, and you're good to go!
You don't need paid Embage plan to build ai agent. On the free plan, you can simply plug in your own API keys and run your agents directly without any AI credit markups.
Embage is still early. If you spot any bugs or need a feature, let us know, I’ll patch or build it as fast as possible.
Give Embage a try using code "LAUNCH_FREE" to claim free AI credits and build your first agent. Excited to hear your feedback!
@tehreem_fatima5 Thank you Tehreem, would love to have your feedback.
Very cool, with all of the connections, how does Embage prevent prompt injection?
@jake_strack Hey Jake, all of the system prompt stays in the server only, we are using cloudflare agents. Currently we are not checking each conversation but will soon add prompt injection score for each message and add threshold for the score, to further reduce injection attacks.
The part that stands out to me is the evaluation layer. Most agent builders stop at 'it answers questions,' but automatically scoring every interaction and pulling structured data out of the transcripts is what makes this workable for a one-person company — you can't sit and read every chat. Custom Datastores fit that too: what the agent collects lands in actual tables instead of a transcript graveyard. Do the evaluation scores feed back into how the sub-agents route conversations, or is that purely reporting for now?
@jcbuilds Hey JC, not only evaluation score but you can add any other information out of transcripts, like you can ask agent to rate it out of 5 so you can actually see stars of each conversation or ask post conversation analysis agent to extract what was the knowledge gap of the conversation so you can add it to your knowledgebase.
But Currently post conversation analysis subagent only run for main agent and not subagents.
Hey Everyone, One thing I want to share is that, you have to avail free AI credits using "LAUNCH_FREE" coupon if you are on free-tier to get AI credits, otherwise you wont be able to build Agents since Free tier by default don't include any AI credits.
Congrats on the launch Aradhya, you deserved it. All the best!
@german_merlo1 Thank you so much German, appreciate it.
The human handoff piece is the part I'd want to stress-test before trusting this with real customers. What happens on the customer's end when the agent requests a handoff but no human is actually online, do they sit in a queue, get a message to wait, or does the agent keep trying to resolve it solo? Running voice agents myself, that gap is usually where the experience actually breaks.
@galdayan Thanks for the question Gal, for human handoff, once the agent thinks human is needed (as per the instruction), till the human takes over, AI chats with the user, and when human takes over AI stops answering till human agent handoff control to AI Agent. From the human agent's perspective, if they are not available, all the requests sits right there till they take over.
@aradhya_tiwari2 good to know the agent doesn't just go silent. one thing we ran into with voice agents though - the AI continuing to chat is fine, but if the customer has no idea whether a human is actually coming, they bail anyway. uncertainty kills it more than the wait itself. is there any signal to the customer about expected wait, or an escalation that pages a human if the queue sits too long?
@galdayan Hey Gal, currently there is no queue, all the handoff request will be visible in one inbox per platform, so that one human operator can chat with multiple end-users, hence there is no queue indicator for users.
Letting the free plan run on your own API keys is a smart way to get people trying it. Do the Telegram and website agents share one datastore or does each get its own?
@neerajnatHi Neeraj, telegram and agent can have same datastore or different, everything is powered by subagents, so it is upto you how you connect it with your subagent.