Geiger Grey

Grey works inside a project. Open your organization in Dash and choose Grey on one of its projects. Everything you create in Grey belongs to the project you have open: its knowledge bases, pipelines, keys, conversations and settings. Everyone in the project's organization can open it, with a role based on their organization role.

Grey opens on the project's Overview. The sidebar lists its areas, from Overview and Knowledge Bases down to Usage and Settings. Sections your role can't see aren't shown, and you can hide the ones you don't use under Settings → Navigation.

To jump to any screen, use Search Grey in the top bar, or press Ctrl K (⌘ K on Mac). If you have no projects yet, Grey shows No projects yet with Create project: name the project and Grey switches to it.

A new project needs a few things in place before it can answer questions. In order:

1. Connect your model provider accounts in Pipelines → Model Providers: at least one that can create embeddings and one with chat models.
2. Pick a processing model in Settings → General, so Grey can build the knowledge graph and describe images.
3. Create a knowledge base in Knowledge Bases → All Knowledge Bases, with an embedding model and an ingestion preset.
4. Add sources to it: upload files, import from Assets and Notes, or add web pages.
5. Create a pipeline in Pipelines → All Pipelines and select the knowledge base under its Retrieval settings.
6. Ask it questions in Playground → Chat.
7. Publish the pipeline, then serve it through an API key, an MCP server or a widget.

Connecting providers, picking the processing model and setting up API keys, MCP servers or widgets need an Owner or Admin. Managers can do the other steps.

Each section of the sidebar covers one part of running Grey, and each has its own guide. The sections on this page sum them up and link to each one.

A knowledge base holds the documents, web pages, notes, images, audio and video Grey answers from. Each source is ingested in the background: read, split into chunks and embedded with the knowledge base's embedding model, which is fixed when you create it. An ingestion preset decides how sources are chunked and enriched.

Add sources by uploading files, importing from Geiger Assets and Notes, or adding web pages. Track them on Sources and Ingestion Jobs, and mark them verified to keep them fresh.

The Knowledge bases and Sources and ingestion guides cover both.

While it ingests your sources, Grey can pick out the people, organizations, products and other entities they mention, and how they relate. They form one knowledge graph per project, which you explore in Graph Explorer, edit and merge in Entities, and shape in Ontology. Pipelines can draw on it when a question is about how things connect.

The Knowledge graph guide covers building and editing it.

A pipeline answers questions from your knowledge bases. It finds the passages that match a question, then has a chat model from your own provider write the answer, with citations. You choose its model, knowledge bases, retrieval settings and prompt, and publish numbered versions.

The Playground is where you try a pipeline: chat with it, see the passages behind each answer, and rate answers. It runs the pipeline's saved settings, so you can test changes before you publish them.

See Pipelines and model providers and Playground.

A published pipeline can answer outside Grey in three ways: an OpenAI-compatible API you call with an API key, an MCP server for AI clients such as Claude Desktop and Cursor, and a chat widget you embed on your website. All three use the pipeline's latest published version.

The API keys, MCP servers and widgets guide covers setting each one up.

• Evaluations score a pipeline against sets of test questions for faithfulness, relevance, precision and recall, and Feedback gathers the answers people rated or marked unanswered. See Evaluations and feedback.
• Usage & Costs shows what Grey recorded as cost by day and kind, with an optional monthly soft cap, and Request Logs traces every question asked. See Usage and request logs.
• Overview sums it all up: queries, spending, ingestion, top pipelines and what needs attention. See Project overview.

Each Grey project has its own roles: Owner, Admin, Manager, Member and Viewer, plus any you create. Settings → Team & Members manages who has which, and Roles & Permissions decides what each role can do. See Team and roles.

Settings → General holds the project's processing and transcription models and query logging, Navigation your own sidebar, and Data Export JSON downloads of the project's data. See Project settings.

• Ctrl K or Ctrl / (⌘ K or ⌘ / on Mac) opens Search Grey, to jump to any screen you can see
• In Chat, Enter sends a message, Shift Enter starts a new line and Esc stops a reply
• In fields that take a list, like aliases, chat models or allowed domains, Enter or , adds each item, and Backspace in the empty field removes the last one