Hydra DB releases Open Glean, a self-hosted AI workspace with 622 GitHub stars in five days

Hydra DB released Open Glean on 17 September 2026, an open-source AI workspace built over its own vector database. The repository, published under Apache-2.0, reached 622 stars in the five days since release, according to the GitHub API.

Open Glean is a Next.js server that answers questions across a user's files, notes, saved webpages and connector-synced knowledge, and writes each answer with inline citations to the sources it retrieved. The project's README describes one composer for asking questions and a Deep Research mode that plans a directed graph of sub-questions, runs each level against Hydra in parallel, writes a finding per branch, and deduplicates sources into one numbered citation list before writing the answer.

How it works

The Next.js server acts as a typed proxy in front of the @hydradb/sdk package, calling api.hydradb.com for retrieval and any OpenAI-compatible endpoint for the answer text. A model picker with OpenRouter favourites is built in, and the project's README lists a OPEN_GLEAN_ALLOW_PRIVATE_LLM_URL flag that permits a local endpoint such as Ollama or LM Studio on a loopback address. Conversations persist to MongoDB when reachable, with a browser fallback.

Hydra DB says in the README that API keys are held server-side in an AES-256-GCM-encrypted, httpOnly session cookie rather than in the browser, and that a stored key is pinned to the endpoint it was stored against. The stated reason is to prevent a caller from pairing an attacker-controlled URL with the server's key and having the server send it there in an Authorization header.

What it takes to run

Open Glean does not include a retrieval engine of its own. It requires a Hydra DB account and API key, or a deployment-level shared HYDRA_API_KEY, to answer any question. A user who already runs a different vector store cannot swap Hydra out through configuration.

The stack is Next.js 16, React 19, Tailwind CSS v4, and the Hydra SDK, with no other runtime data or AI dependency. A Docker image ships the Next.js standalone output as a non-root user, and the repository's NOTICE file records that connector logos are third-party trademarks and are not covered by the Apache grant.

For a team that has been using a hosted knowledge search product, the practical read is that the answer surface is now something they can host, extend and audit, while the retrieval backend is still a paid vendor. That is a partial answer to the audit and residency questions a hosted product raises, and the trade-off is worth stating out loud rather than treating an Apache licence as the whole story.

Source

• Open Glean repository, Hydra DB, published 17 September 2026