feat: add Ollama and AWS Bedrock AI assistants

[singhankur]

Summary

This PR adds two new AI assistant implementations to complement the existing OpenAI assistant:

• OllamaAIAssistant — integrates with locally hosted Ollama models via langchain-ollama. Useful for air-gapped or on-prem deployments that can't use cloud APIs.
• BedrockAIAssistant — integrates with AWS Bedrock via langchain-aws (ChatBedrock). Uses the standard boto3 credential chain (IAM roles, environment variables, ~/.aws/credentials) — no custom proxy or auth headers needed.

Both follow the existing plugin pattern: drop in the class and register it in plugins/ai_assistant_plugin/__init__.py.

Files changed

File	Description
querybook/server/lib/ai_assistant/assistants/ollama_assistant.py	New OllamaAIAssistant class
querybook/server/lib/ai_assistant/assistants/bedrock_assistant.py	New BedrockAIAssistant class
plugins/ai_assistant_plugin/__init__.py	Usage examples for both assistants
requirements/ai/langchain.txt	Add langchain-ollama and langchain-aws

Configuration

Ollama

Set the OLLAMA_BASE_URL environment variable (defaults to http://localhost:11434) and enable the assistant in your plugin:

from lib.ai_assistant.assistants.ollama_assistant import OllamaAIAssistant
ALL_PLUGIN_AI_ASSISTANTS = [OllamaAIAssistant()]

Config example (in Querybook admin settings):

{
  "default": {
    "model_args": { "model_name": "llama3.2" }
  }
}

AWS Bedrock

Requires valid boto3 credentials (IAM role or ~/.aws/credentials):

from lib.ai_assistant.assistants.bedrock_assistant import BedrockAIAssistant
ALL_PLUGIN_AI_ASSISTANTS = [BedrockAIAssistant()]

Config example:

{
  "default": {
    "model_args": {
      "model_id": "anthropic.claude-3-5-haiku-20241022-v1:0",
      "region_name": "us-east-1"
    }
  }
}

Supported models include Amazon Nova, Anthropic Claude 3.x, Meta Llama 3, and Mistral families.

Test plan

• Run Querybook locally with Ollama server running and verify query generation works end-to-end
• Run Querybook with AWS credentials and verify Bedrock assistant responds correctly
• Confirm existing OpenAI assistant is unaffected
• Check that langchain-ollama and langchain-aws install cleanly alongside existing deps