Customer support agents as a service
Train it on your docs in minutes, embed it in one line. It greets returning customers with their history, takes real action in your stack - refunds, tickets, escalations - learns from every resolution, and can provably forget anyone who asks.
Memory by Cognee Cloud knowledge graphs
1 line
to install the widget
0 keys
no LLM keys to manage
2 graphs
knowledge + customer memory
100%
provable right-to-forget
Support bots answer the ticket. Then forget the customer.
Your customers remember. Now your support does too.
Every support bot on the market is stateless retrieval over docs. The customer who wrote in three weeks ago is a stranger again today. EverDesk gives your agent a permanent, queryable memory of every customer relationship - and the hands to act on it, not just talk about it.
What it does
What a stateless support bot cannot do, out of the box.
recall
A customer comes back days later, in a fresh session, and the agent picks up where you left off: 'Yes, we spoke on July 4 about error ZEN-42.'
act
Configure actions in plain English - refunds, tickets, escalations. The agent fires them mid-conversation through your webhooks, with validated parameters and a signed payload, then remembers what it did.
improve
Mark a ticket resolved and the verified fix joins the knowledge graph. The next customer with that problem gets the answer in seconds.
forget
One click hard-deletes a customer's memories from graph and vector store. Their memory graph visibly drops to zero nodes.
graph
Your docs become entities and relationships. Customers become nodes linked to their issues, plans, and resolutions - and you can see it.
ops
Sessions with transcript, tokens, cost, and feedback, plus a live memory feed. Sign in with Google and share the workspace with your team.
Judge view
From a real session on 2026-07-04: one customer's issue becomes another customer's instant answer, then gets forgotten on request.
cognee memory ops
RECALL recall for cust_2b88ecea - I keep hitting error ZEN-42, help? (7578ms)
REMEMBER remember cust_2b88ecea - QA turn stored in memory graph
RECALL recall for cust_2b88ecea - Have we spoken before? What was my issue? (9298ms)
RESOLVE resolution learned - Widget bluetooth pairing fails on Linux
RECALL recall for cust_200db187 - bluetooth pairing fails on my Linux machine, any fix?
REMEMBER remember cust_200db187 - QA turn stored in memory graph
FORGET forgot cust_2b88ecea - 4 memory items hard-deleted from the graph
RECALL recall for cust_2b88ecea - I keep hitting error ZEN-42, help? (7578ms)
REMEMBER remember cust_2b88ecea - QA turn stored in memory graph
RECALL recall for cust_2b88ecea - Have we spoken before? What was my issue? (9298ms)
RESOLVE resolution learned - Widget bluetooth pairing fails on Linux
RECALL recall for cust_200db187 - bluetooth pairing fails on my Linux machine, any fix?
REMEMBER remember cust_200db187 - QA turn stored in memory graph
FORGET forgot cust_2b88ecea - 4 memory items hard-deleted from the graph
Core architecture
Everything your agent knows lives in isolated knowledge graphs on Cognee Cloud.
Website widget
embed.js
Your product
REST API
EverDesk agent
recall -> answer -> act -> remember
Knowledge graph
docs + learned resolutions
Customer memory graph
every customer, every issue
Cognee Cloud
graph + vector, isolated per dataset
The difference
| Stateless support bot | EverDesk | |
|---|---|---|
| Returning customer | A stranger every time | Greeted with full history |
| Resolved tickets | Knowledge lost in logs | Learned by the agent, reused forever |
| Taking action | Suggests; a human does the rest | Fires your webhooks in-conversation |
| Delete my data | Good luck with the logs | Provable graph deletion |
| Understanding | Embedding similarity | Entities and relationships |
| Setup | Weeks of integration | Docs in, script tag out - minutes |
Integrate
A widget for your site, an API for your product. Same agent, same memory.
<script
src="https://everdesk.allensaji.dev/embed.js"
data-everdesk-key="pk_yourcompany_xxxxxxxx"
async>
</script>One script tag before </body>. That is the whole install.
FAQ
In Cognee Cloud knowledge graphs. Each company gets two isolated graph databases: one for its knowledge base, one for customer memory. Every conversation turn becomes graph entities and relationships, not just embeddings.
No. Answers are generated server-side by the memory layer from your graph context. You bring documentation, not API keys.
Both. Beyond answering, you give it actions in plain English - refunds, tickets, escalations - and it triggers them through your webhooks (Zapier, Make, n8n, or a custom endpoint) with schema-validated parameters and HMAC-signed payloads. It never invents an action or its parameters; you declare both. And it remembers every action it took.
Every memory written for a customer is a tracked data item. Forgetting hard-deletes those items from the graph and vector store, and you can verify it: the customer's memory graph visibly drops to zero nodes.
Yes. The widget is one integration; the same agent is exposed as a REST API you can call from your product, mobile app, Slack bot, or backend. See the docs for the full reference.
Only what you provide: pasted text, page URLs, or uploaded files. No repo access, no crawling you did not ask for.
When your team marks a conversation resolved, the verified solution is written into the knowledge graph and becomes a playbook. The next customer with the same problem gets the answer immediately.
Yes. Sign in with Google and invite teammates by email. Every member gets full access to the workspace and its customer memory; owners manage the team.
Docs to a deployed agent - one that remembers and acts - in minutes.
Onboard your company