llng-assistant

LemonLDAP::NG maintenance assistant powered by a local LLM.

Designed for operators who rarely deal with incidents — when your SSO breaks at 11 PM on a Friday, this tool helps you diagnose and fix it without having to read through the documentation.

How it works

You (natural language)
  → Local LLM (Mistral Nemo / Qwen 2.5)
    → llng-mcp (MCP server instances)
      → Multiple LLNG servers (SSH, Docker, Kubernetes, API)

The LLM runs entirely on your workstation. Nothing leaves your network.

Supports managing multiple LLNG instances simultaneously — compare configurations, search sessions across servers, or troubleshoot one instance while monitoring another.

The documentation index (built from LLNG source RST files) is embedded in the Docker image and updated with each release.

Requirements

• Linux or macOS workstation with 8+ GB RAM (16+ GB recommended for 12b models)
• Node.js 20+
• Ollama — local LLM runtime
• Network access to your LLNG servers (SSH, kubectl, or HTTP API)

Installation

1. Install Ollama

Ollama runs LLMs locally on your machine. Install it:

curl -fsSL https://ollama.com/install.sh | sh

Start the server (runs on http://localhost:11434):

ollama serve

2. Download an LLM

Choose a model based on your available RAM:

Model	Download	RAM needed	Quality
qwen2.5:7b	4.7 GB	8 GB	Good
mistral-nemo:12b	5.1 GB	16 GB	Better (recommended)
qwen2.5:14b	8.1 GB	16 GB	Best

ollama pull mistral-nemo:12b

3. Install llng-assistant

git clone https://github.com/linagora/llng-assistant.git
cd llng-assistant
npm install
npm run build

4. Configure your LLNG instances

Create ~/.config/llng-assistant/config.yaml:

llm:
  model: mistral-nemo:12b

mcp_config: ~/.llng-mcp.json

Then configure your LLNG instances — see Configuration below.

5. (Optional) Build the documentation index

This enables semantic search over the LLNG documentation:

ollama pull nomic-embed-text
npm run build-index -- --src /path/to/lemonldap-ng/doc/sources/admin

Docker installation (alternative)

curl -fsSL https://raw.githubusercontent.com/linagora/llng-assistant/main/install.sh | bash

The script handles Docker, Ollama, model download, and systemd service setup.

Usage

llng-assistant

Example interactions (in French or English):

> Pourquoi les utilisateurs ne peuvent plus se connecter sur production ?
> Combien de sessions actives sur production ?
> Compare la config OIDC entre production et staging
> Cherche les sessions de dupont sur staging
> Montre-moi les dernières erreurs SAML sur production
> Quelle est la configuration du vhost app.example.com sur staging ?
> Comment configurer l'authentification Kerberos ?

All tools are automatically prefixed with their instance name, so the LLM knows which LLNG server to query. Ask questions about specific instances by name.

Configuration

Settings are stored in ~/.config/llng-assistant/config.yaml after installation.

The LLM model is configured here. LLNG instances are configured separately via llng-mcp.

Minimal config

llm:
  model: mistral-nemo:12b   # auto-selected based on available RAM

mcp_config: ~/.llng-mcp.json

Then configure your LLNG instances in ~/.llng-mcp.json (standard MCP format):

{
  "mcpServers": {
    "production": {
      "command": "npx",
      "args": ["-y", "llng-mcp"],
      "env": {
        "LLNG_MODE": "ssh",
        "LLNG_SSH_HOST": "sso.example.com",
        "LLNG_SSH_USER": "admin"
      }
    },
    "staging": {
      "command": "npx",
      "args": ["-y", "llng-mcp"],
      "env": {
        "LLNG_MODE": "k8s",
        "LLNG_K8S_CONTEXT": "staging-cluster",
        "LLNG_K8S_NAMESPACE": "auth",
        "LLNG_K8S_POD_SELECTOR": "app=lemonldap-portal"
      }
    }
  }
}

Inline instances (alternative)

Instead of mcp_config, you can define instances directly:

llm:
  model: mistral-nemo:12b

instances:
  production:
    command: npx
    args: ["-y", "llng-mcp"]
    env:
      LLNG_MODE: ssh
      LLNG_SSH_HOST: sso.example.com
      LLNG_SSH_USER: admin

  staging:
    command: npx
    args: ["-y", "llng-mcp"]
    env:
      LLNG_MODE: k8s
      LLNG_K8S_CONTEXT: staging-cluster
      LLNG_K8S_NAMESPACE: auth
      LLNG_K8S_POD_SELECTOR: "app=lemonldap-portal"

For detailed llng-mcp configuration options, see the llng-mcp documentation.

Documentation index

The assistant includes semantic search over the LLNG documentation, powered by nomic-embed-text embeddings via Ollama. The pre-built index is included in Docker images and updated with each release. See step 5 above to rebuild it from source.

Related projects

This tool is built on top of:

• llng-mcp — MCP server for LLNG diagnostics
• LemonLDAP::NG — The excellent SSO platform we maintain

License

AGPL-3.0 — Copyright LINAGORA