Skip to content

Latest commit

 

History

History
52 lines (40 loc) · 2.61 KB

README.md

File metadata and controls

52 lines (40 loc) · 2.61 KB

Vertex AI Gemini Pro Slack bot example

A Slack chat bot that uses the Gemini 1.0 Pro Vision model. It can accept images and text from Slack.

Deploying

  1. Navigate to Slack settings, Tools & Settings > Manage apps.
  2. Click on Build in the top right corner.
  3. Click on Create new app and select From manifest.
  4. Copy the contents slack-manifest.yaml into the text box.
  5. Proceed with creation of the Slack application and install it to your workspace.
  6. You should now be able to view application credentials. Take the App ID and Signing Secret and place them in terraform.tfvars (slack_app_id and slack_signing_key).
  7. Next click on OAuth & Permissions and take the Bot User OAuth Token and place it in terraform.tfvars (slack_token).
  8. You can now run terraform apply to deploy the bot. After deployment, you should see the Cloud Run Json2Pubsub URL in the output.
  9. In Slack app settings, click on Event Subscriptions and replace the Request URL with the URL from the previous step.
  10. In Slack, create a channel and add the bot to it (channel settings, Integrations > Add an app).
  11. You can now discuss with the bot via private messages or tagging it in a channel!

Integrating Vertex AI Search

By setting vertex_search parameter, you can deploy an additional Gemini function call which can search your Vertex AI Datastore. Follow the Vertex AI Search quickstart (which uses Alphabet's financial documents) and enable it in terraform.tfvars:

vertex_search = {
  enabled      = true
  location     = "eu" # or global, us
  datastore_id = "YOUR_DATASTORE_ID_12345678" # replace this with your datastore ID
}

Once deployed, Gemini is given the possibility to function call the Vertex Search function and integrate the results.

Please note that for now, the multimodal search functionality (eg. images) will not work in conjunction with the function calling. The model also might be less efficient at answering common knowledge questions related in any way to finance (work in progress!).

Example Terraform connfiguraiton

Example terraform.tfvars file:

project_id        = "<your-project-id>"
region            = "europe-west1"
slack_token       = "xoxb-...." # Slack OAuth token for workspace
slack_signing_key = "8e................02f" # Slack signing key
slack_app_id      = "A........L" # App ID

See the previous article: Building an AI Slack Bot using Vertex Generative AI