allen-saji/everdesk

EverDesk is a customer support chat agent designed to remember individual customers across conversations instead of treating every chat as a fresh start. A company trains it on their own documentation, embeds a single script tag on their website, and the agent can then answer customer questions using that knowledge while also recalling what happened with that specific customer before. When a returning customer starts a new conversation, the agent recognizes them and can refer back to earlier issues rather than asking them to explain everything again. When a support conversation is marked resolved, the fix that worked gets written into a shared knowledge base, so the next customer with the same problem gets an immediate answer instead of waiting for a person to solve it again. The agent can also take actions on a company's behalf, such as filing a refund ticket or notifying a team, by calling webhooks that the company sets up in plain English descriptions, with safety checks like rate limits and signed requests to prevent misuse. Customers can also ask to be forgotten, which deletes their stored memory completely and is reflected immediately in the company's dashboard. Behind the scenes, EverDesk uses a memory service called Cognee Cloud to store two separate types of data for each company: their knowledge base documents and a shared graph of customer memories. It uses Groq to decide which action, if any, should fire during a conversation. Companies can use the hosted version directly by signing up and dropping in the embed script, or developers can self-host it. Self-hosting needs Node.js 20 or newer, a Cognee Cloud account, an Upstash Redis database, and Google sign-in credentials for the dashboard, with a Groq API key optional for the action-taking features. The project is written in TypeScript and released under the MIT license, and it currently has 0 stars on GitHub.