Multi-agent Production

Agentic Production Environment

A LangGraph-based multi-agent framework for chat-style customer support: one surface for users, a supervisor that routes to specialist agents, and production hooks from APIs through containers to observability.

View on GitHub Back to commercial experience
Diagram: router and specialist agents

Overview

Clients interact through a single chat experience while LangGraph directs messages to domain specialists—support, billing, technical, escalation—each equipped with retrieval, tools, and MCP integrations. Conversation history is available over GraphQL for analytics dashboards; synchronous chat uses REST. Infrastructure assets include Docker and Kubernetes manifests plus scripts for repeatable builds and rollout.

Capabilities

Supervised routing

Central router keeps orchestration explicit and extensible as new intents or policies are introduced.

RAG and tooling

Agents combine grounded answers with tool calls and MCP servers for live systems access.

Operational APIs

REST endpoint for messaging plus GraphQL schema for sessions and turns to power reporting products.

Human in the loop

Documentation covers HITL patterns so sensitive escalations can be handed to people cleanly.

Observability

Guidance for RAG evaluation (including RAGAS), runtime faithfulness checks, and Langfuse-style tracing.

Deployment artefacts

Dockerfile, Kubernetes namespace, deployment, service, HPA, and helper scripts under the infra folder in the repository.

Visual summary

Diagram: REST and GraphQL surfaces, packaging and documentation
REST and GraphQL surfaces plus packaging and documentation map. The router-and-specialist overview is shown in the hero above.

Technology stack

  • LangGraph
  • LLM APIs
  • REST
  • GraphQL
  • MCP
  • Docker
  • Kubernetes

Building a similar assistant platform?

Get in touch