OpenIngest is open source — and it's yours to run.

The document AI engine that ran on Tensorlake for over a year is now fully open source. As we go all-in on agent infrastructure, the hosted Document AI product winds down on June 30 — and OpenIngest is how it lives on.

Author

Shanshan Wang
June 17, 2026


Tensorlake

Product Update · June 2026

OpenIngest is open source — and it's yours to run.

The document AI engine that ran on Tensorlake for over a year is now fully open source. As we focus on agent infrastructure, the hosted Document AI product winds down on June 30 — and OpenIngest is how it lives on.

This month.OpenIngest — the document AI engine ran on Tensorlake — is now open source, and the README takes you from clone to fully self-hosted. As we flagged last month, the hosted Document AI product retires on June 30; self-hosting OpenIngest is the path forward. On Sandboxes — import any Docker image, public or private, and run it as a sandbox, plus support for Claude managed agents. Read the documentation to get started.

OpenIngest · Now open source

Document AI is Open Source, now it's fully yours to own and run.

OpenIngest is the document AI engine that parses, extracts, and structures real-world documents — the same engine that ran in production on Tensorlake orchestration for over a year. It's now open source. Follow the README and you can self-host it 100%, on your own infrastructure, with full control of your data.

Get the code on GitHub →

Action needed · Document AI sunset

The hosted Document AI product retires on June 30, 2026.

We're focusing on agent infrastructure, so the hosted Document AI service is winding down. On June 30 the hosted API stops responding and API keys are deactivated, and the Document AI docs come down with it. We gave a month's notice last issue — this is the final reminder.

What to do: move to OpenIngest. It's the same engine, open source, and self-hostable end to end — the README is the migration guide. If you're running document workloads on the hosted API today, clone the repo and stand up your own deployment before June 30.

Start here: github.com/tensorlakeai/OpenIngest

Sandboxes · Docker image import

Bring any Docker image — public or private.

Import a registry image and run it as a sandbox — no rebuild required. We pull the layers and write them straight to the sandbox filesystem, converting your container image into a VM image you can launch. Works with Docker Hub, GHCR, Amazon ECR, Google Container Registry, Quay, and self-hosted registries; private images authenticate through your existing docker login credentials, so it drops right into CI. Register the image once, then launch sandboxes from it in seconds.

See importing images in our docs.

Agents · Claude Managed Agents

Claude's brain, your sandbox's hands.

Anthropic runs the LLM, the agent loop, and session state on its own infrastructure, while every tool call executes inside your own Tensorlake sandbox — you control the execution environment and the data. Suspended sandboxes resume in well under a second for tight loops, snapshots capture filesystem and process state, and you can fork a checkpoint into parallel sandboxes to explore several solutions at once. Run the orchestrator your way: webhook-in-sandbox, polling, or webhook.

Read the Claude Managed Agents guide.

Try Sandboxes

We're offering free credits to developers who try Tensorlake Sandboxes and share their feedback. Whether you're importing your own Docker images, running Claude managed agents, or scaling parallel workloads, we want to hear what you build and what we can improve.

Get Started Today

Check out our docs page to get started

Tensorlake · AI Infrastructure for Agents