hydra-db/open-glean

选用指南

AI 根据 README 整理 · 2026/10/14 13:36:17

使用方式

通过 Next.js 服务器运行，提供 Web 界面，用户通过浏览器访问，支持 Docker 部署。

查看 README 依据

- **Bring your own model**: any OpenAI-compatible endpoint. A searchable OpenRouter model picker with favourites is built in. ## Getting started ```bash npm ci npm run dev ``` Copy `.env.example` to `.env.local` and set at least `OPEN_GLEAN_SESSION_SECRET`. Everything else is optional. Without a Hydra key, the app shows a connect screen and asks for one. Open http://localhost:3000, then: 1. **Connect**: paste your Hydra DB API key (create one at [app.hydradb.com/keys](https://app.hydradb.com)) and pick a database.

2. **Add a model**: in Settings, paste an OpenRouter (or other OpenAI-compatible) key and pick a model. It only writes answers; search works without it. 3. **Ask away**, or add context first from the Context page (files, notes, URLs). ## Running with Docker ```bash docker build -t open-glean . docker run -p 3000:3000 \ -e OPEN_GLEAN_SESSION_SECRET=your-random-16-char-secret \ -e HYDRA_API_KEY=your-hydra-key \ -e OPENROUTER_API_KEY=your-openrouter-key \ -e OPEN_GLEAN_LLM_MODEL=google/gemini-3.7-flash \ -e MONGODB_URI=mongodb://your-mongo-host:27017 \ open-glean ```

部署要求

需要 Node.js 环境（npm ci）或 Docker；需要设置 OPEN_GLEAN_SESSION_SECRET，可选配置 Hydra API 密钥、OpenRouter API 密钥和 MongoDB URI。

查看 README 依据

- **Bring your own model**: any OpenAI-compatible endpoint. A searchable OpenRouter model picker with favourites is built in. ## Getting started ```bash npm ci npm run dev ``` Copy `.env.example` to `.env.local` and set at least `OPEN_GLEAN_SESSION_SECRET`. Everything else is optional. Without a Hydra key, the app shows a connect screen and asks for one. Open http://localhost:3000, then: 1. **Connect**: paste your Hydra DB API key (create one at [app.hydradb.com/keys](https://app.hydradb.com)) and pick a database.

2. **Add a model**: in Settings, paste an OpenRouter (or other OpenAI-compatible) key and pick a model. It only writes answers; search works without it. 3. **Ask away**, or add context first from the Context page (files, notes, URLs). ## Running with Docker ```bash docker build -t open-glean . docker run -p 3000:3000 \ -e OPEN_GLEAN_SESSION_SECRET=your-random-16-char-secret \ -e HYDRA_API_KEY=your-hydra-key \ -e OPENROUTER_API_KEY=your-openrouter-key \ -e OPEN_GLEAN_LLM_MODEL=google/gemini-3.7-flash \ -e MONGODB_URI=mongodb://your-mongo-host:27017 \ open-glean ```

已知限制

DocumentDB 不支持 IAM 数据库认证；未设置 OPEN_GLEAN_LLM_MODEL 时答案不可用；允许私有 LLM URL 默认关闭。

查看 README 依据

| `HYDRA_DEFAULT_DATABASE` | Optional default tenant. Requests with no explicit database scope to this one. | | `MONGODB_URI` | Chat persistence. A MongoDB connection string, or a bare DocumentDB cluster endpoint (auto-detected via `.docdb.*.amazonaws.com`). | | `MONGODB_DB` | Database name for chats (defaults to `open_glean`). | | `OPENROUTER_API_KEY` | Optional server-level LLM key. When set, users need not enter their own. | | `OPEN_GLEAN_LLM_MODEL` | Default answer model id, e.g. `openai/gpt-4o-mini`. No built-in default. Without it (or a per-user model), answers are unavailable. |

| `OPEN_GLEAN_LLM_BASE_URL` | Optional OpenAI-compatible endpoint for the server-level key (defaults to OpenRouter) | | `OPEN_GLEAN_ALLOW_PRIVATE_LLM_URL` | Allow an LLM base URL on a private or loopback address, for a local model such as Ollama or LM Studio. Off by default. It permits plaintext `http:` on a private address only. A public host still requires `https:`, and non-http schemes are rejected. | | `MONGODB_PROXY_KEY` | Shared secret for the Lambda proxy. The Lambda reads the same value as `PROXY_KEY`; the two names must match. |

| `OPEN_GLEAN_MAX_CONCURRENT_RESEARCH` | Concurrent Deep Research runs per instance (default 3). Each run costs many LLM and retrieval calls | > **DocumentDB does not support IAM database authentication.** The IAM token > path in the code cannot authenticate against Amazon DocumentDB. `AWS_REGION`, > `AWS_IAM_USER_ARN` and `AWS_ACCESS_KEY_ID` / `AWS_SECRET_ACCESS_KEY` are read > by that path and are listed here only to describe what it expects. For > DocumentDB, use a standard MongoDB connection string or the Lambda proxy. ## Stack