Open source, vendor-agnostic enterprise AI · Setup guide

Gebo.ai setup guide

Seven steps from an empty server to your users chatting with the company documents, then the integrations with your systems and with other AI tools, the monitoring, the single sign-on with the users and groups, and the chats and agents. Step 3b is an alternative to step 3.

Download the complete guide (PDF)

  1. 1Install Gebo.ai with DockerPDF

    Download the reference Docker Compose file and start Gebo.ai with the four databases it uses. Two commands, then a few minutes of downloading.

  2. 2First access: create the administratorPDF

    Open Gebo.ai in a browser and create the administrator account. Two minutes.

  3. 3Connect the AI models (OpenAI)PDF

    The easy setup wizard connects Gebo.ai to an AI provider: one API key configures the chat model and the model that indexes your documents. This page uses OpenAI; step 3b does the same with Regolo.ai.

  4. 3bConnect the AI models (Regolo.ai)PDF

    The same wizard as step 3 with Regolo.ai, the inference service of Seeweb S.r.l. that serves open models from European data centres through an OpenAI-compatible API. Use this page instead of step 3.

  5. 4Create a knowledge base and a projectPDF

    A knowledge base groups the documents your users chat with. Its projects hold the data sources the documents come from.

  6. 5Connect the company folder and publish itPDF

    Let Gebo.ai read a company folder, create the data source on it and publish it: Gebo.ai reads and indexes the documents.

  7. 6Schedule the evening syncPDF

    Make Gebo.ai check the company folder every evening, so the index follows the folder without anyone publishing by hand.

  8. 7Chat with your documentsPDF

    Create an account for a user, then chat as that user with the company documents.

Integrations

Each page shows where the credentials of a third-party system go in Setup; adding the data source then works as in step 5. The last pages connect Gebo.ai with other AI tools and agents over MCP and A2A, and cover the API keys they use.

  1. 8ConfluencePDF

    Give Gebo.ai an Atlassian account that reads your Confluence spaces. Setup keeps the credentials; a data source then chooses the spaces, as the company folder of step 5.

  2. 9JiraPDF

    Give Gebo.ai an Atlassian account that reads your Jira Cloud projects: issues, comments and attachments become searchable.

  3. 10SharePoint and OneDrivePDF

    Let Gebo.ai read your SharePoint sites and OneDrive libraries through an application registered in Microsoft Entra, or a SharePoint 2019 server with a user of its own.

  4. 11Google Drive and WorkspacePDF

    Let Gebo.ai read shared drives and folders of your Google Workspace through a service account that acts on behalf of a user.

  5. 12AWS S3 and S3-compatible storagePDF

    Let Gebo.ai read documents kept in S3 buckets, on AWS or on any S3-compatible store (MinIO, Ceph, Wasabi, a NAS).

  6. 13WebDAV: Nextcloud, ownCloud, Synology and othersPDF

    Let Gebo.ai read a WebDAV share: Nextcloud, ownCloud, OpenCloud, Seafile, Pydio Cells, ONLYOFFICE Workspace, Synology DSM, IIS or Apache WebDAV.

  7. 14Uploaded files and git repositoriesPDF

    Two data sources need nothing in Setup: files uploaded by hand, and git repositories whose credentials live in the data source itself.

  8. 15Use external MCP serversPDF

    Connect a server that speaks the Model Context Protocol: its tools become functions the chat models and the agents of Gebo.ai can call.

  9. 16Expose Gebo.ai as an MCP serverPDF

    Publish Gebo.ai’s tools, knowledge bases and prompts as an MCP server for Claude Desktop, Cursor, IDE extensions and automations. Every client acts with the rights of the user who connects, through a personal API key.

  10. 17Import A2A agentsPDF

    Register an agent that speaks the Agent2Agent protocol, published by another platform or by another Gebo.ai: it becomes a participant of your networks of agents.

  11. 18Expose Gebo.ai agents over A2APDF

    Publish an agent or a whole network of agents as an Agent2Agent endpoint: other platforms, and other Gebo.ai installations, call it as an opaque agent with a card of its own.

  12. 19Administrative API keysPDF

    Keys an administrator generates on behalf of a user, for scripts, services and integrations that call Gebo.ai, its MCP servers and its A2A endpoints without a person logging in.

Monitoring

Where an administrator sees how the models and the ingestion are used, what each ingestion run did, and who is signed in.

  1. 20LLM usage and costsPDF

    How much each model is used, for whom and by which part of Gebo.ai, how fast it answers and, once the models have prices, what it costs.

  2. 20bProvider deals and model pricesPDF

    The costs of page 20 come from the prices of the models. Gebo.ai imports them from the providers that publish them; for the others, OpenAI among them, enter them here.

  3. 21Ingestion trafficPDF

    How many documents the ingestion received, processed, passed on, left out or failed, and how many chunks and tokens it produced, for each knowledge base, project and data source.

  4. 22Ingestion runs and their logsPDF

    Every publish and every scheduled sync of a data source is a run. The Logs tab lists the runs of all the data sources, with their outcome and the log of each.

  5. 23Active sessionsPDF

    Who is using Gebo.ai, since when, from where and with what: logins, API keys and tokens of the single sign-on providers. Any of them can be ended at once.

Single sign-on, users and groups

Your people sign in with the identity provider they already use: one page per provider, then the users and the groups that decide who gets in and what they see.

  1. 24Microsoft Entra IDPDF

    Your people sign in to Gebo.ai with their Microsoft work account. Register Gebo.ai as an application of your Entra tenant, then enter it in Setup.

  2. 24bMicrosoft accountsPDF

    People sign in to Gebo.ai with a personal Microsoft account (Outlook.com, Hotmail, Live) or with the work account of the organizations you list. One application in Entra, one configuration in Setup.

  3. 25GooglePDF

    Your people sign in to Gebo.ai with their Google Workspace account. Create an OAuth client in the Google Cloud console, then enter it in Setup.

  4. 26AWS CognitoPDF

    Your people sign in to Gebo.ai with the accounts of an Amazon Cognito user pool. Add Gebo.ai as an app client of the pool, then enter it in Setup.

  5. 27KeycloakPDF

    Your people sign in to Gebo.ai with the accounts of a Keycloak realm. Add Gebo.ai as a client of the realm, then enter it in Setup.

  6. 28Other OpenID Connect providersPDF

    Any identity provider that speaks OpenID Connect, such as Okta, Auth0, Authentik or Zitadel: you enter its addresses yourself, from its discovery document.

  7. 28bSingle sign-on: platform settingsPDF

    Two settings of the platform’s configuration file shape single sign-on: whether the people your provider recognizes become users on their own, and how your own applications call Gebo.ai with the tokens of your provider.

  8. 29Users and groupsPDF

    Who can sign in and how: the accounts Gebo.ai keeps itself and those of the single sign-on providers, gathered in groups that open knowledge bases and projects to some people only.

Chats and agents

What your people chat with: the chat profiles, the deep search across your sources, and the networks of agents that answer.

  1. 30Chat profilesPDF

    A chat profile is what your people chat with: the knowledge bases it searches, the models it uses and who may use it. Gebo.ai starts with one that searches every knowledge base a person can see; add others for a team or a topic.

  2. 30bR.a.g. autotunePDF

    How close a document fragment must be to a question for the chats to use it is the similarity threshold. The right value depends on the embedding model and on your documents: R.a.g. autotune measures it on your own documents and keeps it current as they change.

  3. 31Deep searchPDF

    For one question, deep search reads the documents of several sources, your knowledge bases, the web and your company systems, and answers with a report that quotes them. As administrator you set the sources it reads and who may search them.

  4. 32Networks of agentsPDF

    The chats are answered by networks of agents. Gebo.ai ships ready networks for the chats with knowledge bases and for the free chats: choose among them, or build your own from a copy.