# Windows PowerShell
$env:YANDEX_API_KEY = "AQVNxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
$env:YANDEX_FOLDER_ID = "b1gxxxxxxxxxxxxxxxxx"

# Permanent (survives restart)
[System.Environment]::SetEnvironmentVariable('YANDEX_API_KEY', 'AQVNxxx...', 'User')
[System.Environment]::SetEnvironmentVariable('YANDEX_FOLDER_ID', 'b1gxxx...', 'User')

# Linux/Mac
export YANDEX_API_KEY="AQVNxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
export YANDEX_FOLDER_ID="b1gxxxxxxxxxxxxxxxxx"

# Add to ~/.bashrc for permanent
echo 'export YANDEX_API_KEY="your_key"' >> ~/.bashrc
echo 'export YANDEX_FOLDER_ID="your_folder"' >> ~/.bashrc

llm:
  provider: yandex

  yandex:
    api_key: ${YANDEX_API_KEY}
    folder_id: ${YANDEX_FOLDER_ID}
    model: "qwen3-235b-a22b-fp8/latest"
    temperature: 0.7
    max_tokens: 2000
    timeout: 180

python -c "
from src.llm.yandex_provider import YandexProvider
from src.llm.base_provider import LLMConfig

config = LLMConfig(provider_type='yandex')
provider = YandexProvider(config)
print(f'Provider: {provider}')
print('Connection successful!')
"

gpt://<folder_id>/<model_name>
emb://<folder_id>/<embedding_model>

gpt://b1g123456789/qwen3-235b-a22b-fp8/latest
gpt://b1g123456789/yandexgpt/latest
emb://b1g123456789/text-search-doc/latest

import os
from openai import OpenAI

# Initialize client with Yandex endpoint
client = OpenAI(
    api_key=os.getenv("YANDEX_API_KEY"),
    base_url="https://llm.api.cloud.yandex.net/v1",
    default_headers={
        "x-data-logging-enabled": "false",  # Privacy
        "x-folder-id": os.getenv("YANDEX_FOLDER_ID"),
    },
)

# Construct model URI
folder_id = os.getenv("YANDEX_FOLDER_ID")
model_uri = f"gpt://{folder_id}/qwen3-235b-a22b-fp8/latest"

# Make request
response = client.chat.completions.create(
    model=model_uri,
    messages=[
        {"role": "system", "content": "You are a code analysis expert."},
        {"role": "user", "content": "Explain MVCC in PostgreSQL."},
    ],
    temperature=0.7,
    max_tokens=2000,
)

print(response.choices[0].message.content)

# Stream response
stream = client.chat.completions.create(
    model=model_uri,
    messages=[
        {"role": "user", "content": "List security best practices for C code."},
    ],
    stream=True,
)

for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

# Get embeddings
embedding_uri = f"emb://{folder_id}/text-search-doc/latest"

response = client.embeddings.create(
    model=embedding_uri,
    input="Find buffer overflow vulnerabilities",
    encoding_format="float",
)

embedding = response.data[0].embedding
print(f"Embedding dimension: {len(embedding)}")  # 256

from src.llm.yandex_provider import YandexProvider
from src.llm.base_provider import LLMConfig

# Initialize provider (reads from config.yaml)
config = LLMConfig(
    provider_type='yandex',
    temperature=0.7,
    max_tokens=2000,
    extra_params={
        'api_key': os.getenv('YANDEX_API_KEY'),
        'folder_id': os.getenv('YANDEX_FOLDER_ID'),
        'model': 'qwen3-235b-a22b-fp8/latest',
    }
)
provider = YandexProvider(config)

# Generate response
response = provider.generate(
    system_prompt="You are a PostgreSQL security expert.",
    user_prompt="Find SQL injection vulnerabilities in the executor module."
)

print(response.content)
print(f"Tokens used: {response.metadata['tokens_used']}")
print(f"Latency: {response.metadata['latency_ms']:.0f}ms")

from src.workflow.orchestration.copilot import Copilot

# Copilot automatically uses configured provider
copilot = Copilot()
result = copilot.answer("Find memory allocation functions in PostgreSQL")
print(result['answer'])

llm:
  # Set Yandex as the active provider
  provider: yandex

  yandex:
    # Required: Authentication
    api_key: ${YANDEX_API_KEY}      # Service account API key
    folder_id: ${YANDEX_FOLDER_ID}  # Yandex Cloud folder ID

    # Model selection
    model: "qwen3-235b-a22b-fp8/latest"  # Best for code analysis

    # API endpoint (default, usually not changed)
    base_url: "https://llm.api.cloud.yandex.net/v1"

    # Timeouts
    timeout: 180  # seconds (increased for large prompts)

    # Generation parameters
    temperature: 0.7       # Creativity (0.0-1.0)
    max_tokens: 2000       # Max response length
    top_p: null            # Use Yandex default

    # Embedding model
    embedding_model: "text-search-doc/latest"

default_headers={
    "x-data-logging-enabled": "false",  # Disables logging on Yandex side
    "x-folder-id": folder_id,
}

Error: 401 Unauthorized

# Check if variables are set
echo $YANDEX_API_KEY
echo $YANDEX_FOLDER_ID

# Verify API key format (should start with AQVNw, AQVN_, or similar)
python -c "import os; key = os.getenv('YANDEX_API_KEY', ''); print(f'Key prefix: {key[:5]}...' if key else 'NOT SET')"

# Verify API key scope includes yc.ai.languageModels.execute

Error: 429 Too Many Requests

# Built-in retry handles this automatically
# But you can increase timeout for complex requests
yandex:
  timeout: 300  # 5 minutes

import time

for attempt in range(3):
    try:
        response = provider.generate(system_prompt, user_prompt)
        break
    except YandexRateLimitError:
        time.sleep(2 ** attempt)  # Exponential backoff

Error: Request timeout

# Increase timeout for large prompts/responses
yandex:
  timeout: 300  # seconds

Error: Folder not found or access denied

# Automatic retry for:
# - APITimeoutError (timeout)
# - APIConnectionError (connection issues)

# Retry parameters:
# - max_retries: 3
# - initial_delay: 2.0 seconds
# - max_delay: 30.0 seconds
# - backoff_factor: 2.0