[Bug] Google Gemini provider: AttributeError 'GenerativeModel' has no attribute 'bind_tools' in Docker image v1.0.0

[RonnyAllen]

When using CellAtria with PROVIDER="Google" and a Gemini API key via the Docker image
ghcr.io/astrazeneca/cellatria:v1.0.0, the container starts but immediately crashes with:

AttributeError: 'GenerativeModel' object has no attribute 'bind_tools'.

It looks like the Google / Gemini integration in this image is using a bind_tools method that is not present on the current GenerativeModel object from the Google GenAI SDK.

Environment

• CellAtria image: ghcr.io/astrazeneca/cellatria:v1.0.0
• Run method: Docker Desktop on Windows 11
• Machine: local laptop
• LLM provider (in .env): PROVIDER="Google"
• Google model: GOOGLE_MODEL=gemini-2.5-pro
• Using free‑tier Gemini API key from Google Cloud

.env configuration (redacted)

#########################################

General Settings (Required for all)

#########################################

Specifies which LLM provider to use.

Supported options: "Azure OpenAI", "OpenAI", "Anthropic", "Google", "Local"

PROVIDER="Google"

#########################################

Azure OpenAI Configuration

Required if PROVIDER="Azure OpenAI"

#########################################

Azure OpenAI API key

AZURE_OPENAI_API_KEY=

Azure OpenAI endpoint (e.g., https://.openai.azure.com/)

AZURE_OPENAI_ENDPOINT=

The deployment name of your Azure model (e.g., gpt-4o)

AZURE_OPENAI_DEPLOYMENT_NAME=

Azure OpenAI API version (e.g., 2024-03-01-preview)

AZURE_OPENAI_API_VERSION=

sampling / generation parameters ---

AZURE_OPENAI_TEMPERATURE=
AZURE_OPENAI_TOP_P=

#########################################

OpenAI Configuration

Required if PROVIDER="OpenAI"

#########################################

Your OpenAI API key (starts with "sk-...", get from https://platform.openai.com/)

OPENAI_API_KEY=

The model to use (e.g., gpt-4, gpt-3.5-turbo)

OPENAI_MODEL=

#########################################

Anthropic Configuration

Required if PROVIDER="Anthropic"

#########################################

Your Anthropic API key (starts with "sk-ant-...", get from https://console.anthropic.com/)

ANTHROPIC_API_KEY=

#########################################

Google Gemini / Vertex AI Configuration

Required if PROVIDER="Google"

#########################################

Google Gemini API key (get from Google Cloud Console)

GOOGLE_API_KEY=

Google Gemini model name (e.g., gemini-pro)

GOOGLE_MODEL=gemini-2.5-pro

The .env file is being read successfully now (no more “.env file not found” errors).
The crash only happens when CellAtria tries to construct the LLM agent with tools.

Docker run / volumes
I am running the container via Docker Desktop UI, but the equivalent CLI is:
docker run -it --rm
-p 7860:7860
-v "C:\Users\rohan\OneDrive\Documents\TIFR PhD\Mahendra Sonawane Lab\Papers":/data
-v C:\cellatria-env:/mnt/work/projects
ghcr.io/astrazeneca/cellatria:v1.0.0
cellatria

Here, C:\cellatria-env is a junction pointing to my local cellatria-main checkout containing the .env file, so inside the container the env file is at:
/mnt/work/projects/.env

This fixed the earlier .env file not found at: /mnt/work/projects/.env error.

Full traceback
From the container logs:
Traceback (most recent call last):
File "/usr/local/bin/cellatria", line 27, in
graph, cellatria = create_cellatria(args.env_path)
File "/opt/cellatria/agent/base.py", line 57, in create_cellatria
llm_with_tools = llm.bind_tools(tools)
AttributeError: 'GenerativeModel' object has no attribute 'bind_tools'

Expected behavior
Using PROVIDER="Google" and a valid GOOGLE_API_KEY / GOOGLE_MODEL should start the CellAtria app successfully and allow normal tool-enabled conversations.

Actual behavior
Container exits with the AttributeError above before the UI becomes available.

Questions / requests

1. Is Gemini currently supported in ghcr.io/astrazeneca/cellatria:v1.0.0?
2. If yes, is there a specific GOOGLE_MODEL or SDK version that is required?
3. If this is a known incompatibility with newer Google GenAI SDKs, would it be
   possible to:

• Pin to a compatible SDK version in the image, or

• Update the code to use the current recommended pattern for tool calling with Gemini (e.g., model = GenerativeModel(..., tools=...) or equivalent)?

Thanks a lot for releasing CellAtria — I’m trying to use it for a zebrafish development / metabolism PhD project and would love to get the Google provider path working on the free tier.