Helix Foundry on Railway
HelixDB's open-source app for turning company data into a graph with an ontology, on Railway.
Helix Foundry is an open-source application from HelixDB that connects company data sources, such as database tables and documents, into one graph with an ontology. You can ask questions in natural language and receive answers that cite where they came from, and the app exposes an MCP endpoint so other AI tools can query the same graph.
It runs on HelixDB, the project's graph-vector database, and uses an AI provider of your choice for drafting the ontology and answering questions. The project is young and its versions are in the 0.1 range, so expect change between releases.
Helix Foundry, HelixDB and the bundled ontology agent are released under Apache-2.0.
What Helix Foundry can do
- Graph built from tables and documents
- Ontology drafting with an AI model
- Cited answers to natural-language questions
- MCP server for AI tools
- HelixDB graph database
- Acme Corp sample data to try it
What people use it for
- Combining data from several systems into one queryable graph
- Question answering over company documents with citations
- Giving AI assistants a structured view of company data through MCP
- Evaluating graph-based knowledge tooling with sample data
Videos
Running Helix Foundry on Railway
The Railway template is a community template, not affiliated with HelixDB. It deploys seven services: a Caddy gateway that asks for a generated password, the Foundry web app, the Foundry backend, HelixDB, SeaweedFS as S3-compatible storage, MySQL and an optional Postgres with the sample data, with volumes for state and images pinned by digest. The template notes say Foundry needs about 1 to 2 GB while drafting or loading data, and synced data counts as egress.
Helix Foundry templates on Railway
Tracklayer templates are listed first. Other Helix Foundry templates follow, ordered by total deployments on Railway. How this works.
Deployment numbers as of 2026-11-20.
Questions about Helix Foundry
How do I log in?
Open the gateway URL and sign in with BASIC_AUTH_USERNAME (default admin) and BASIC_AUTH_PASSWORD, both under the gateway variables. Then pick an AI provider under Settings, AI models, or set ANTHROPIC_API_KEY or OPENAI_API_KEY on the Foundry service beforehand.
What drives the cost?
Seven services run, and Foundry needs about 1 to 2 GB of memory while drafting or loading, so Hobby or Pro is recommended. Synced data counts as egress because Foundry reaches its sources over the network.
Is data kept across redeploys?
Yes, Foundry, HelixDB, storage and MySQL use volumes. Keep FOUNDRY_CONNECTOR_KEY safe, because without it stored connector credentials cannot be read.
Can I run more than one replica?
No. Keep every service at one replica because workers, HelixDB and the store keep state in-process or on volumes. Scale vertically instead.
Are the MCP routes protected by the password?
The MCP and HTTP Ask routes skip the password prompt and need a bearer token. Set FOUNDRY_AGENT_ENDPOINTS to off to keep them behind the password.
What licence applies?
Helix Foundry, HelixDB and SeaweedFS are Apache-2.0. MySQL is GPL-2.0 with the Universal FOSS Exception.
Similar apps
Tracklayer maintains the templates marked as such; Helix Foundry is a trademark of its owners and this site is not affiliated with them.