Hardened RAGFlow Per-Instance Secrets on Ubuntu 24.04

This product provides RAGFlow v1.0.0 on Ubuntu 24.04 (x86_64) with seller-written software that changes how the stack is exposed and which credentials it uses. The upstream quick-start configuration publishes every backend service on all network interfaces, uses the same well-known password for every service, uses a default administrator password and leaves self-registration on. This AMI keeps the upstream images unmodified and changes the configuration around them. The additional charges cover seller-performed integration, configuration and testing on the bounded baseline described below. Any support or maintenance offered by the seller is a supplementary service and is not the sole basis for the software charges.

What the additional charges cover:

• Internal-only backends: Elasticsearch, MySQL, object storage, Kvrocks and NATS are not published to the network at all. Only the web entry on ports 80 (redirect) and 443 is published; the RAGFlow API and admin ports are bound to 127.0.0.1.
• Per-instance secrets: on first boot every password in the stack's environment file, including the administrator password, is replaced with a random value. The file is readable only by root.
• Self-registration is switched off. Users are created with the hanwei-ragflow command.
• A per-instance self-signed TLS certificate (EC P-256) covering the instance's addresses, served with upstream's HTTPS configuration.
• All six images pinned by digest, so the running software is exactly what was tested.
• A verifier with 20 checks across operating system settings, RAGFlow settings and licence files, including live probes, with a JSON report. It runs hourly and reports drift; it does not revert changes.
• Testing of the above on the stated baseline.

Technical implementation:

• A compose override file (docker-compose.hanwei.yml) is applied on top of a per-instance copy of the upstream docker directory; the pristine upstream tree is kept separately and never edited.
• Services are started by ragflow-stack.service. sshd is configured for public-key login only; kernel settings include vm.max_map_count required by Elasticsearch.

Included components:

• Ubuntu 24.04.5 LTS (x86_64), Docker Engine 29.1.3, Docker Compose 2.40.3
• RAGFlow v1.0.0 (Apache-2.0), unmodified
• Elasticsearch 8.11.3 (Elastic License 2.0), MySQL 8.0.40 (GPL-2.0), Silo RELEASE.2026-08-06 (AGPL-3.0), Kvrocks 2.16.0 (Apache-2.0), NATS 2.14.2 (Apache-2.0), unmodified images
• Hanwei RAGFlow Guard 1.0.0 (seller software)

Tested deployment baseline:

• Instance type and hardware: t3.xlarge, 4 vCPU, 16 GiB memory, 100 GiB gp3 root volume. The idle stack used about 6.5 GB of memory.
• Operating system and architecture: Ubuntu 24.04.5 LTS, x86_64.
• Functional checks: first boot on two separate instances, the 20-check verifier, a port scan from a second instance showing only ports 22, 80 and 443 open, differing secrets and certificates between instances, administrator and user sign-in, dataset creation and file upload, reboot persistence, and detection of a deliberately introduced default password.
• Scope boundary: document parsing, indexing, retrieval and chat were not tested; no model provider was used. Connectors, MCP servers, agents, web interface screens, other instance types, performance, scale, high availability, backups and upgrades were not tested.

Security and operational boundaries:

• Open TCP 22 and TCP 443 only from your own address range, or reach the web interface through an SSH tunnel. The certificate is self-signed.
• You supply the embedding and chat model providers and their credentials. Prompts and documents you send go to the providers you configure.
• RAGFlow blocks requests from its model, connector and tool features to private addresses (ALLOW_ANY_HOST is 0, upstream describes 1 as development-only). A model server inside your VPC is not reachable unless you change that setting.
• Elasticsearch is licensed under the Elastic License 2.0, which does not allow offering it to others as a hosted or managed service. MySQL is under GPL-2.0 and Silo under AGPL-3.0. Licence files, digests and a written source offer are in /usr/share/doc/hanwei-ragflow/THIRD-PARTY-IMAGES.md.
• Data is stored in Docker volumes on the root disk. No backups, high availability or in-place RAGFlow updates are included.

Suitable use cases:

• Running a single-node, self-hosted RAGFlow with a model provider of your choice, where backend services should not be reachable from the network.

Pricing statement:

• Marketplace software charges cover the integrated software configuration and tested deployment baseline described above. AWS infrastructure charges and third-party model charges are billed separately.

RAGFlow is a separate open-source project of InfiniFlow. The seller is not affiliated with, endorsed by, or sponsored by InfiniFlow.