Work/Usher AI
Our own product
In developmentUsher AI: Support That Acts
An open-source, self-hostable AI customer support platform. Embed it on a website or in a mobile app; it answers from your docs, acts on your live systems through your own MCP server, and escalates to a human when it is not confident.
- Node/TypeScript
- Mastra
- Postgres + pgvector
- MCP
- Docker
The idea
Embed it on a website or in a mobile app. It answers from an organisation's own docs via RAG, reaches into their live backend data through their own MCP server, and escalates to a human when it is not confident.
Why it is different
The closest open-source helpdesks — Chatwoot, Tiledesk — bolt AI onto a ticketing system. Usher AI starts from an agent with first-class MCP tool-calling, so it can act on live systems mid-conversation rather than only reciting static content.
Design principles
- Simple by default, enterprise when needed — docker compose up plus an LLM key gets a working bot; SSO, RBAC, phone and voice are optional and off by default
- Provider-agnostic for LLM, speech-to-text/text-to-speech, telephony and identity
- Docker-first from the first commit
- Stateless app so scaling is additive
Tech
- Node/TypeScript with Mastra as the agent framework
- Postgres + pgvector as the only datastore
- MCP client for live tool-calling
- RAG ingestion for docs, FAQs and policies
- Docker Compose with a bundled mock MCP server so tool-calling works out of the box
Status
Usher AI is still in development. It is MIT-licensed and built to be self-hosted; the repository is private today.