Stars
Forks
Watchers
Developer links
AgentDesk
Most AI chat integrations bolt an LLM onto a text box and call it customer support; AgentDesk builds the entire helpdesk around AI agents that handle first responses, escalate to humans when they cannot answer, and track follow-up work through integrated ticket workflows. Each AI agent binds to a specific LLM configuration, one or more knowledge bases, a set of callable Skills, and MCP tool integrations, creating specialized support personas for different product lines or departments. When a customer message arrives, the agent retrieves relevant FAQ entries and document chunks through RAG-powered vector search (Qdrant), then passes retrieved context through an Answerability Gate that determines whether the knowledge base can support a reliable answer; if not, it triggers a fallback response and recommends human handoff instead of hallucinating. Human agents pick up escalated conversations in a dedicated workspace that preserves the full AI conversation history, customer profile, conversation tags, and linked tickets, so they never start from scratch. The ticket system converts conversations into tracked work items with category assignment, status flows, progress notes, and team routing. An admin dashboard provides user management, role-based access control, team scheduling, and automatic conversation assignment rules. Model configuration supports any OpenAI-compatible provider, letting you wire in LLMs, embedding models, and reranking models with per-model context limits, timeout settings, and rate controls. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. Apache 2.0 licensed.
Benefits
- AI-First With Reliable Human Escalation
- AI agents handle initial responses using RAG retrieval and an Answerability Gate that triggers human handoff when knowledge confidence is low, instead of hallucinating answers.
- Specialized Agent Personas Per Product
- Each AI agent binds to specific LLM configs, knowledge bases, Skills, and MCP tools, letting you create separate support personas for different departments or product lines.
- Conversation-to-Ticket Workflow Loop
- Convert any support conversation into a tracked ticket with category assignment, status flows, progress notes, and team routing without losing the original chat context.
- Any OpenAI-Compatible Model Provider
- Configure LLMs, embedding models, and reranking models from any OpenAI-compatible API with per-model context limits, timeout settings, retry behavior, and rate controls.
Features
- Knowledge Base RAG
- Store FAQs, documents, and retrievable content with automatic chunking and Qdrant vector indexing. Retrieval logs and quality analysis track answer accuracy over time.
- Agent Workspace
- Support agents view conversation lists, handle AI-to-human handoffs, reply to customers, transfer to teammates, link customer profiles, and manage ticket context.
- Answerability Gate
- Before responding, the system evaluates whether retrieved content can support a reliable answer. Low-confidence queries trigger fallback messages and human escalation.
- Ticket Management
- Create tickets from conversations, assign categories and teams, move through configurable status flows, record progress notes, and close the resolution loop.
- Model Configuration
- Register OpenAI-compatible LLMs, embedding models, and rerankers with API endpoints, keys, context limits, output caps, and independent enable/disable toggles.
- Skills and MCP Tools
- Extend AI agents with callable Skills for business logic and MCP protocol integrations for connecting to external enterprise systems during conversations.