Gebo.ai setup guide · Step 5 of 7

Connect the company folder and publish it

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

Time
5 minutes, plus the indexing time of your documents.
You need
The addresses of the company shares and an account that can read them. With Docker, also the time zone of your office.

1On the server: give Gebo.ai the company shares

How Gebo.ai reaches the company shares depends on how it is installed: with Docker Compose (step 1), as a service on a Linux virtual machine, or on a Windows server. Follow the option that matches your installation.

Option 1: Gebo.ai in Docker Compose

Gebo.ai reads the company documents from /opt/gebo.ai/shares inside its container. Mount each company share, of a Windows file server or a NAS, in its own subfolder there, as a Docker volume of type cifs or nfs. Docker mounts it like an fstab line, with the options you write, and Gebo.ai sees it read only.

Write the lines in a file named docker-compose.override.yml in the gebo-ai folder of step 1, next to docker-compose.yml. Docker Compose reads both files, so docker-compose.yml can be downloaded again on upgrades without losing your shares. The first volume line keeps the volume of step 1 on /opt/gebo.ai/shares but without :ro: Docker creates the subfolders in it, which a read-only volume does not allow.

services:
  gebo.ai:
    environment:
      TZ: "Europe/Rome"
    volumes:
      - "gebo-shares:/opt/gebo.ai/shares"
      - "finance:/opt/gebo.ai/shares/finance:ro"
      - "hr:/opt/gebo.ai/shares/hr:ro"
volumes:
  finance:
    driver: local
    driver_opts:
      type: cifs
      device: "//files/Finance"
      o: "addr=files,user=${SMB_USER},pass=${SMB_PASSWORD},iocharset=utf8,ro"
  hr:
    driver: local
    driver_opts:
      type: nfs
      device: ":/export/hr"
      o: "addr=nas,nfsvers=4,ro"

Write the account that reads the Windows shares in a file named .env in the same folder, readable only by you (chmod 600 .env). Docker Compose fills ${SMB_USER} and ${SMB_PASSWORD} from it:

SMB_USER=gebo-reader
SMB_PASSWORD=...

Restart Gebo.ai with the shares (the databases keep running) and list what it sees:

docker compose up -d
docker compose exec gebo.ai ls /opt/gebo.ai/shares
Changing a share laterDocker keeps the options a volume was first created with. After changing them, run docker compose down, remove the volume, for example docker volume rm gebo-monolith_finance, then docker compose up -d. This removes only the mount: the files stay on the file server.
Set the time zoneThe TZ line sets the time zone of the scheduled sync of step 6. Without it the container runs on UTC, and an evening sync at 20:00 would start at 21:00 or 22:00 Italian time.

Option 2: Gebo.ai as a service on a Linux virtual machine

The Linux appliance of gebo.ai/downloads and the Ubuntu installation of the technical configuration manual (section Linux Ubuntu and Debian style installation) both install Gebo.ai from the .deb package, which runs it as the gebo-ai service under the gebo-ai user.

Mount each Windows share on the virtual machine through /etc/fstab, read only and owned by that user, in one parent folder, for example /mnt/company:

sudo apt install cifs-utils language-pack-en
sudo locale-gen en_US.UTF-8
sudo update-locale LANG=en_US.UTF-8
sudo mkdir -p /mnt/company/finance
# /etc/gebo-share, chmod 600: username=..., password=..., domain=...
sudo nano /etc/fstab

Add one line per share to /etc/fstab, then mount it and check that the service user can read it:

//files/Finance /mnt/company/finance cifs ro,credentials=/etc/gebo-share,uid=gebo-ai,gid=gebo-ai,iocharset=utf8,_netdev 0 0
sudo mount -a
sudo -u gebo-ai ls /mnt/company/finance

With NFS the line is files:/export/finance /mnt/company/finance nfs ro,_netdev 0 0; the files must then be readable by the gebo-ai user on the NFS server. Finally add the parent folder /mnt/company in Gebo.ai, as described in option 3.

Option 3: Gebo.ai on a Windows server

The Windows installation is the .msi package of gebo.ai/downloads, described in the manual’s section Installing Gebo.ai on a windows server. On Windows, and for the Linux service of option 2, you choose the shared folders in the Gebo.ai setup. In Setup → Enterprise integrations, click Shared filesystems paths setup, then Add folder. In Add a root folder of a shared filesystem choose, among the server’s disks and folders, the one holding the company documents, enter a Description and save. Repeat for each disk or folder.

A Windows share must be visible to the Gebo.ai service as a disk or folder of the server. The installer creates the Gebo.ai Service under the Local System account, which does not see the drive letters mapped by a logged-in user: make the share available to the service, and readable by the account it runs as.

Set the character setOn Linux, run the server in a UTF-8 locale and keep iocharset=utf8 in the options of a Windows share, in the Docker volume or in /etc/fstab. They keep the file and folder names with accented letters and other non-English characters as they are; the manual also documents codepage=cp850.

2In Gebo.ai: create the data source

In Admin → Knowledge bases & projects, click +Add data source next to the project of step 4 and choose +filesystem.

The data source types.
The data source types.

Enter a Description, for example Company shared folder (1), then click the pencil (2) to choose the folders.

The new filesystem data source.
The new filesystem data source.

Expand the shared folder (shares with Docker, or the folder added in the setup), tick the subfolders to publish and click OK.

Choosing the folders: the subfolders, not the shares root.
Choosing the folders: the subfolders, not the shares root.
Tick the subfoldersTicking the subfolders keeps the knowledge base safe when a share is momentarily unreachable: the run stops with an error and the documents already read stay in Gebo.ai.
Choose the folders firstA data source keeps the folders chosen when it is saved: to read other folders, add another data source.

3Publish

The chosen folders appear in the editor (1). Click Publish! (2).

The data source, ready to publish.
The data source, ready to publish.

The Publishing log opens. Gebo.ai lists the documents, reads them, indexes them for semantic search and for full-text search. When the status reads Processed completely, your users can chat with the documents. You can close the window: publishing goes on in the background.

The publishing log: 5 documents received and processed at every stage, none in error.
The publishing log: 5 documents received and processed at every stage, none in error.
Changes are picked up by themselvesGebo.ai also watches the chosen folders and indexes new and changed files within about a minute. The scheduled sync of step 6 is the nightly safety net that checks the whole folder.