Environment Variables Required for Each LLM Provider in interviewstreet/hiring-agent

The interviewstreet/hiring-agent repository requires three specific environment variables—DEFAULT_MODEL, LLM_PROVIDER, and GEMINI_API_KEY—to configure Large Language Model providers, with only GEMINI_API_KEY being mandatory when using Google Gemini models.

The interviewstreet/hiring-agent open-source project centralizes its LLM configuration through environment-based settings read at runtime. Understanding what environment variables are required for each LLM provider ensures proper authentication and model routing without unexpected fallbacks.

Core Configuration Variables

The application reads all LLM settings in prompt.py using Python's os.getenv method. These variables determine which model executes your prompts and whether API credentials are needed.

DEFAULT_MODEL

The DEFAULT_MODEL variable specifies which model name the agent invokes during inference. It maps directly to provider-specific model identifiers.

• Environment Variable: DEFAULT_MODEL
• Default Value: "gemma3:4b" (defined as DEFAULT_MODEL_NAME in the source)
• Provider Impact: Used by both Ollama and Gemini providers to identify the specific model weights

LLM_PROVIDER

The LLM_PROVIDER variable identifies which backend serves the requests. This determines the code path taken and which authentication checks run.

• Environment Variable: LLM_PROVIDER
• Default Value: "ollama" (derived from ModelProvider.OLLAMA.value)
• Valid Options: Values must match the ModelProvider enum defined in models.py (e.g., "ollama", "gemini")

GEMINI_API_KEY

The GEMINI_API_KEY variable stores your Google API credential. This is the only secret required among the environment variables, and it is mandatory exclusively for Gemini deployments.

• Environment Variable: GEMINI_API_KEY
• Default Value: Empty string ""
• When Required: Only when LLM_PROVIDER is set to "gemini"

How Configuration Loading Works in prompt.py

The central configuration logic resides in prompt.py, where the application validates provider selections and applies defaults:

# prompt.py (excerpt)

DEFAULT_MODEL = os.getenv("DEFAULT_MODEL", DEFAULT_MODEL_NAME)
PROVIDER = os.getenv("LLM_PROVIDER", DEFAULT_PROVIDER.value)

# Validate provider

if PROVIDER not in [p.value for p in ModelProvider]:
    PROVIDER = DEFAULT_PROVIDER.value

# API key for Gemini

GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "")

The code first attempts to read each variable from the environment, falling back to hard-coded defaults when variables are missing. It then validates the provider string against the ModelProvider enum imported from models.py, reverting to "ollama" if an invalid value is supplied.

Practical Configuration Examples

Below are concrete implementations showing how to set these environment variables for different deployment scenarios.

Configuring for Gemini

When using Google's Gemini models, you must export all three variables:

# .env file

DEFAULT_MODEL=gemini-2.5-pro
LLM_PROVIDER=gemini
GEMINI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxx

# Python validation

import os
from prompt import DEFAULT_MODEL, PROVIDER, GEMINI_API_KEY

print(f"Using model: {DEFAULT_MODEL}")
print(f"Provider: {PROVIDER}")

if PROVIDER == "gemini":
    assert GEMINI_API_KEY, "GEMINI_API_KEY must be set for Gemini models"
    # Initialize Gemini client...

Running with Default Ollama Setup

If you omit environment variables entirely, the agent defaults to a local Ollama instance:

$ python main.py

# Executes with DEFAULT_MODEL="gemma3:4b" and LLM_PROVIDER="ollama"

No API keys are required for this configuration.

Summary

• DEFAULT_MODEL controls which model weights load, defaulting to "gemma3:4b"
• LLM_PROVIDER selects the backend implementation, defaulting to "ollama" and validated against the ModelProvider enum in models.py
• GEMINI_API_KEY is the only mandatory secret, required exclusively when LLM_PROVIDER=gemini
• All configuration logic is centralized in prompt.py using standard os.getenv calls

Frequently Asked Questions

What happens if I don't set any environment variables?

The application runs using bundled defaults defined in prompt.py: DEFAULT_MODEL becomes "gemma3:4b" and LLM_PROVIDER becomes "ollama". No API keys are required for this local-only configuration.

Is GEMINI_API_KEY required for Ollama models?

No. The GEMINI_API_KEY variable is only evaluated when LLM_PROVIDER resolves to "gemini". When using Ollama (the default), the code ignores this variable entirely, allowing the application to run without any API credentials.

Where are the provider values validated?

Provider strings are validated against the ModelProvider enum in models.py. If the LLM_PROVIDER environment variable contains an invalid value not present in the enum, prompt.py automatically falls back to "ollama" as a safety measure.

Where can I find a template for these variables?

The repository includes an .env.example file at the root level that documents the expected variable names and example values for both Ollama and Gemini configurations. This serves as the authoritative reference for environment setup.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}