Skip to main content

System Overview

The Jeen backend is a distributed system of 13 microservices plus an MCP tool ecosystem. Services are grouped into four layers.

Layer 1: Client-Facing (BFF)​

These services receive requests from the frontend.

ServiceRole
user-serviceAggregates data from downstream services. Handles user profile, sharing, tags, pins, favorites, admin settings, Langflow flows, documents, activity logs. Does not own a database -- delegates all persistence.
auth-serviceAuthenticates users via Firebase or ZITADEL, issues JWT tokens, manages sessions in Redis.

Layer 2: Core Business Logic​

These services implement the platform's AI capabilities.

ServiceRole
llm-coreCentral orchestration. Manages conversations, agents, model/provider catalog, templates, canvas, text conversions, and LLM usage tracking. Calls completion-service for LLM responses and agent-service for agentic execution.
completion-serviceLLM proxy. Routes completion requests to the correct provider (OpenAI, Azure, Anthropic, Google, Mistral, Jamba, Ollama, vLLM, or custom). Handles streaming. Emits token usage events to RabbitMQ.
agent-serviceRuns agentic loops. Receives a message + allowed tools, calls completion-service for LLM reasoning, executes tools via MCP, feeds results back, and repeats until done (max 15 iterations).
document-serviceOrchestrates the full document lifecycle: upload to Azure Blob/S3, trigger parsing via RabbitMQ, trigger embedding via RabbitMQ, track status, serve downloads, manage folders.

Layer 3: Processing Services​

These services do the heavy computational work.

ServiceRole
parser-serviceConverts documents to markdown. Supports 4 parser backends: Azure Document Intelligence, PyMuPDF, MinerU, Marker. Runs an HTTP API and a RabbitMQ worker.
embedding-serviceChunks text, generates vector embeddings (OpenAI/Azure OpenAI), optionally translates to English. Runs an HTTP API and a RabbitMQ worker.
rag-serviceQueries the vector store. Embeds the user query, runs pgvector cosine similarity search, optionally reranks results (BGE model or LLM-based).

Layer 4: Data and Administration​

These services manage persistent data and configuration.

ServiceRole
user-base-msUser data persistence. Stores users, roles, tags, favorites, pins, activity logs, features, shares, locks, integration tools, Langflow accounts, connectors, languages.
admin-base-msOrganization/tenant management. Stores organizations, RBAC (roles, permissions, modules, features, actions), resource units, and per-org configuration (models, agents, connectors, templates, workflows, parsing techniques, languages).
integration-serviceThird-party integration gateway. Manages account provisioning, token lifecycle, and API key management for external tools (currently Langflow).

MCP Ecosystem​

The MCP (Model Context Protocol) layer extends the agent's capabilities with external tools.

ComponentRole
mcp-client-serviceGateway that discovers all MCP tool servers at startup, maintains a tool registry, and routes call-tool requests to the correct server.
7 MCP tool serversIndividual servers exposing tools: document search (RAG), spreadsheet analysis, web search, video/image generation, Python code execution, Atlassian (Jira/Confluence), and interactive UI components.

High-Level Architecture​

                          Frontend / Client Apps
                                  |
                    +-------------+-------------+
                    |                           |
              auth-service                user-service
              (login/register)            (BFF / aggregator)
                    |                           |
                    |             +------+------+------+------+------+
                    |             |      |      |      |      |      |
                    |          llm-core  admin  user  doc-   ident.  langflow
                    |             |      base   base  svc    svc
                    |             |      -ms    -ms
                    |      +------+------+
                    |      |             |
                    |  completion    agent-service
                    |  -service          |
                    |      |        mcp-client-service
                    |      |             |
                    |   LLM APIs    +----+----+----+----+----+----+----+
                    |               | rag|sprd|web |veo2|code|atl.|ui  |
                    |               | tool|sht |srch|    |exec|    |comp|
                    |
              +-----+-----+
              | Firebase   |
              | ZITADEL    |
              | Redis      |
              +------------+

    [document-service] --RabbitMQ--> [parser-service worker]
    [document-service] --RabbitMQ--> [embedding-service worker]
    [completion-service] --RabbitMQ--> [llm-core] (transactions)

    Shared DB (document database): document-service, parser-service,
                                   embedding-service, rag-service