2.3.101 Satellite RAGFlow

RAGFlow

Handle: ragflow
URL: http://localhost:34990

RAGFlow is a document-focused RAG engine with parsing, datasets, hybrid retrieval, and chat/agent workflows. Harbor packages upstream's CPU deployment with Elasticsearch, MySQL, S3-compatible storage, and Valkey. The five data/log volumes persist when the service is stopped. The image is currently published for linux/amd64; ARM64 hosts need an upstream source build.

Starting

Set six unique secrets before first start. The preflight container rejects missing, short, or non-alphanumeric values. harbor config set normally echoes the new value, so redirect its output when setting secrets. Save the secrets securely with the data-volume backups.

harbor config set ragflow.mysql_password "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor config set ragflow.es_password "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor config set ragflow.s3_password "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor config set ragflow.redis_password "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor config set ragflow.admin_password "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor config set ragflow.secret_key "$(openssl rand -hex 32)" >/dev/null 2>&1
harbor up ragflow ollama --open

First startup downloads several large images and initializes the databases. Sign in as admin@ragflow.io with ragflow.admin_password, or register another local account. The bootstrap admin password is used on first creation only; changing its Harbor setting later does not reset the existing account. Keep ragflow.secret_key stable across restarts so sessions remain valid.

For local models, open Model providers → Ollama, set the Base URL to http://ollama:11434, verify and save the instance, then set default LLM and embedding models. The integration ensures Harbor Ollama is healthy and pulls ragflow.chat_model (qwen2.5:3b) and ragflow.embedding_model (mxbai-embed-large). Select those models in RAGFlow's UI; changing the Harbor settings does not automatically change a saved RAGFlow account's model selection. Create a Dataset, choose General parsing, upload a document with Parse on creation, and use Retrieval testing to check its indexed content. The UI can be used with other supported cloud or local model providers too.

The comment mentioning Kotaemon refers to a separate service that Harbor already has; this RAGFlow deployment does not replace it.

Configuration and data

Setting	Default	Purpose
ragflow.image, ragflow.version	infiniflow/ragflow:v0.27.2	RAGFlow CPU image
ragflow.host_port	34990	Web UI and proxied API host port
ragflow.bind_host	127.0.0.1	Host address for the published port
ragflow.es_memory	8g	Elasticsearch container memory limit
ragflow.mysql_password	Empty	Required MySQL root/RAGFlow metadata password
ragflow.es_password	Empty	Required Elasticsearch password
ragflow.s3_password	Empty	Required object-storage password
ragflow.redis_password	Empty	Required Valkey password
ragflow.admin_password	Empty	Required first-run admin password
ragflow.secret_key	Empty	Required persistent session-signing secret
ragflow.register_enabled	1	Set to 0 after creating accounts to disable open registration
ragflow.chat_model	qwen2.5:3b	Ollama model pulled by the integration
ragflow.embedding_model	mxbai-embed-large	Ollama embedding model pulled by the integration

The stack persists ragflow-mysql-data, ragflow-es-data, ragflow-s3-data, ragflow-redis-data, and ragflow-logs as named Docker volumes. Back up all five volumes together with the six secrets. The admin API and dependency ports are not published by Harbor. The web port binds only to loopback by default. For remote users, deliberately change ragflow.bind_host, put TLS and access controls in front of the service, and consider disabling open registration. Do not expose the internal stores directly to the internet.

Troubleshooting

• The UI can answer HTTP 200 before its Python API has completed first-run migrations. Wait for the backend, then reload the page if it initially reports a gateway error.
• Check docker logs harbor.ragflow and the service's ragflow-logs volume if startup or parsing fails. harbor logs tails indefinitely by default.
• Elasticsearch needs enough memory and a sufficiently high vm.max_map_count on the Docker host; follow the upstream quickstart if it cannot start.
• After editing services/ragflow/default.env in a Git checkout, run harbor config update before starting; use harbor config set for local settings rather than editing .env.