Sealed credentials
Passwords and API keys are sealed when you enter them. Your agent never sees them.
An open-source alternative to Palantir Foundry
Connect your databases, SaaS tools and documents. Foundry turns them into one ontology your agents can query, with a source behind every answer.
Foundry is built for a bigger screen: open this page on your computer to start.
Star on GitHub to find it laterConnect your databases, SaaS tools and documents. Foundry turns them into one ontology your agents can query, with a source behind every answer.

Connects to the databases, tools and documents you already run
Ontology
Your app database, Stripe and WorkOS each hold part of every customer. Foundry suggests one ontology that matches the same records across them.
Linked to
How it works
Connect a source, say what you want to know, and review the ontology Foundry drafts before it loads into HelixDB.
Ask
Ask in plain words. Foundry's agent reads your ontology step by step, and its answer cites each record, row and quote it used.
How many support tickets came in each month?
Support tickets grew from 4 in February to 28 in August, and June was the busiest month, with 33[1]. All 119 tickets in the load are counted[2].
Sources
The point
Connect a source and Foundry drafts the ontology. You review and edit it, and every version is kept so you can go back.
Your data, your models
On Helix Cloud, Helix runs Claude for you, or your own Anthropic key on Pro and Max. Self-hosted, use Claude, OpenAI, Azure OpenAI, Amazon Bedrock, Ollama or any OpenAI-compatible server.
Passwords and API keys are sealed when you enter them. Your agent never sees them.
Connectors read your systems and never write back. Later syncs read only what changed.
Through the Helix MCP, an agent has your role, and every change to your setup needs your confirmation.
Self-hosted, Foundry listens on 127.0.0.1 only, and your data stays on your machine unless you pick a hosted model.
Pricing
Plans are per workspace and differ mostly in how often your sources sync. Every plan drafts, loads and answers questions about your ontology.
$0 a month
For trying Foundry on your own data.
$100 a month
For teams that work from fresh data.
$500 a month
For data that has to stay current.
$0 forever
For one person, with Docker. Your data stays on it unless you pick a hosted model.
Your Kubernetes, your cloud or on-prem, with high availability, directory sync and your own models. See Foundry for enterprise.
Each workspace's AI spend is capped each month. Compare plans in the docs, or read them as pricing.md.
Helix Foundry connects your databases, tools and documents, lets an agent draft an ontology over them, and loads it into HelixDB as one connected graph. You can explore it, see where its values came from, and ask questions whose answers cite their sources.
Helix Foundry is open source under Apache 2.0 and starts free, on Helix Cloud or on your own computer. An agent drafts the ontology from the data you connect, so you can start without modelling everything by hand first.
Helix Cloud, at foundry.helix-db.com, is hosted for you, with scheduled syncs and workspaces for your team. For an organisation, Foundry deploys on your own Kubernetes, in your cloud or on-prem, with high availability: see Foundry for enterprise at /foundry/enterprise. The local install is free, for one person on their own computer, and starts with one Docker Compose command. There is no sign-in locally, so never expose it through a proxy, tunnel or port forward.
On Helix Cloud, yes. Each workspace has members with roles, and an agent connected through the Helix MCP acts with your role, so it can only do what you could do on the website. The local install is for one person.
On Helix Cloud, Helix runs Anthropic's Claude models for you, or your own Anthropic key on Pro or Max. On your computer, Foundry uses Claude or OpenAI when it finds an ANTHROPIC_API_KEY or OPENAI_API_KEY, otherwise Ollama, otherwise LM Studio. Settings › AI models switches to Azure OpenAI, Amazon Bedrock or any OpenAI-compatible server.
Docker with Compose v2, ports 3000 and 8333 free, and room for a few GB of images on the first start. Your data lives in Compose volumes, which docker compose down keeps. Never run docker compose down -v unless you mean to delete everything.
Each load writes your ontology into HelixDB, an open-source graph database with native vector and full-text search, as one connected graph. Every value keeps where it came from: a table row, or a quote in a document. HelixDB is Apache 2.0 too.
Helix Cloud runs the same open-source build, so you can connect a source and explore your ontology without standing up HelixDB, DuckDB and Ollama yourself.
Foundry is built for a bigger screen: open this page on your computer to start.
Star on GitHub to find it laterRolling it out to a team? Book a demo