AgentRelAI Assistants Grounded in Your Docs

AgentRel is a platform Jutsu built for AI developer relations, which it called "the AI agent relationship platform": a team connects its documentation and gets an AI assistant that answers developers' questions from it, with sources. Each company gets an organization with its own subdomain and branded sign-in, invites its team with fine-grained permissions, and creates a project per product. A project pulls in GitHub docs repositories (with the docs framework detected automatically), website sitemaps, YouTube channels and uploaded PDF, Word and CSV files. Each ingestion runs as a job that logs every file and the tokens each model used, and the content is chunked and embedded into the project's own Qdrant collection. Answers stream back with cited sources and three suggested follow-up questions, in the dashboard's playground, on the project's public page on agentrel.com and in a chat widget any site can embed, and the dashboard tracks users, messages, response times, feedback and token cost. I built most of it, with 482 of the 518 commits to its main repository: the Bun and Hono API over MongoDB and Qdrant, the plugin server that crawls the sources, the RAG chat on the Vercel AI SDK with GPT-5 mini and Claude Opus 4.1, sign-in with email, Google and GitHub, and most of the React dashboard. The screenshots show it running locally with a demo organization and made-up data.