← all repositories
hydra-db/open-glean

Glean, but you hold the keys

A self-hostable AI workspace over Hydra DB — ask across your files, notes, and connected apps, and get answers that cite their sources.

open-glean
Too new to rank. Velocity needs a few days of history.
not enough history yet
star history

What it does Open Glean is a Next.js chat workspace that treats a Hydra DB database as its memory. One composer retrieves context — memories, files, saved webpages, connector-synced knowledge — and streams an answer with inline citations, a sources panel, and optional web search via the LLM provider’s plugin. You can scope queries to specific databases or collections from the top bar, browse Hydra’s knowledge graph as a mindmap, and run connector syncs without leaving the app. Any OpenAI-compatible endpoint writes the answers; the retrieval itself is Hydra’s job.

The interesting bit Deep Research is the headline act: for questions one query can’t answer, it plans a DAG of sub-questions, runs each level against Hydra in parallel, writes a finding per branch, and dedupes everything into one numbered citation list — a timeline shows the plan and live progress. Nearly as notable is the key hygiene: keys sit in an AES-256-GCM-encrypted, httpOnly cookie, are resolved server-side, and are pinned to the endpoint they were stored with, so a caller can’t pair an attacker-controlled URL with the server’s key. That’s a threat model most side projects never bother writing down.

Key highlights

  • Deep Research: DAG-planned sub-questions, parallel per level, per-branch findings, deduped citations, live progress timeline.
  • Key handling: nothing key-shaped reaches the browser — encrypted httpOnly cookie, server-side resolution, endpoint pinning.
  • Bring your own model: any OpenAI-compatible endpoint, plus a searchable OpenRouter picker; search works without an LLM key, answers don’t.
  • Persistence: chats land in MongoDB (or DocumentDB) when reachable, with a localStorage fallback.
  • Lean stack: Next.js 16, React 19, Tailwind v4, and the typed @hydradb/sdk — no other runtime dependency for data or AI. Apache-2.0, minus the connector logos (third-party trademarks).

Caveats

  • The open-source part is the workspace, not the brain: retrieval, connectors, and the knowledge graph live behind Hydra DB’s hosted API (HYDRA_BASE_URL defaults to api.hydradb.com). Whether Hydra DB itself is self-hostable isn’t stated in the README.
  • Deep Research is priced in tokens — each run makes many LLM and retrieval calls, and concurrency is capped at 3 per instance by default.
  • DocumentDB users hit a known wall: the IAM auth path in the code can’t authenticate against DocumentDB; the README points to a standard connection string or the Lambda proxy instead.

Verdict Worth a look if you want a Glean-style, citation-backed workspace over your own data and are already in Hydra DB’s orbit — or just want to see a DAG-of-sub-questions research loop implemented cleanly. Skip it if “self-hosted” means the retrieval engine too; here, that part is a service.

Frequently asked

What is hydra-db/open-glean?
A self-hostable AI workspace over Hydra DB — ask across your files, notes, and connected apps, and get answers that cite their sources.
Is open-glean open source?
Yes — hydra-db/open-glean is open source, released under the Apache-2.0 license.
What language is open-glean written in?
hydra-db/open-glean is primarily written in TypeScript.
How popular is open-glean?
hydra-db/open-glean has 1.6k stars on GitHub.
Where can I find open-glean?
hydra-db/open-glean is on GitHub at https://github.com/hydra-db/open-glean.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.