bfagent-llm

LLM integration for the BFAgent Hub ecosystem.

Features

• Prompt Framework: DB-driven prompt templates with Jinja2 rendering
• Secure Template Engine: Sandboxed rendering with security validations
• Multi-Layer Caching: L1 (local) + L2 (Redis) + L3 (DB) caching
• Resilient Service: Retry, circuit breaker, tier fallback
• LLM Adapters: Gateway, OpenAI, Anthropic, Fallback

Installation

pip install bfagent-llm
# or with all providers
pip install "bfagent-llm[all]"
# or from git
pip install "bfagent-llm @ git+https://github.com/achimdehnert/platform.git#subdirectory=packages/bfagent-llm"

Quick Start

from bfagent_llm import PromptFramework

# Get singleton instance
framework = PromptFramework.get_instance()

# Execute a prompt template
result = await framework.execute(
    template_code="expert_hub.generate_content",
    context={"topic": "Python async programming"},
    tier="standard",
)
print(result.content)

Components

SecureTemplateEngine

Sandboxed Jinja2 rendering with security features:

• Blocks dangerous patterns (eval, exec, __dunder__)
• Context sanitization
• JSON schema validation
• Custom safe filters

from bfagent_llm import SecureTemplateEngine

engine = SecureTemplateEngine()
result = engine.render(
    template="Hello {{ name }}!",
    context={"name": "World"},
)
print(result.rendered)  # "Hello World!"

PromptRegistry

Multi-layer caching for prompt templates:

from bfagent_llm import CachedPromptRegistry

registry = CachedPromptRegistry()
template = registry.get("expert_hub.generate_content", tenant_id=tenant_id)

ResilientPromptService

LLM calls with resilience patterns:

from bfagent_llm import ResilientPromptService, GatewayLLMAdapter

adapter = GatewayLLMAdapter(base_url="http://llm-gateway:8100")
service = ResilientPromptService(llm_client=adapter)

result = await service.execute(
    system_prompt="You are a helpful assistant.",
    user_prompt="Explain quantum computing.",
    tier="standard",
)

LLM Adapters

from bfagent_llm import (
    GatewayLLMAdapter,    # BFAgent LLM Gateway
    OpenAILLMAdapter,     # Direct OpenAI API
    AnthropicLLMAdapter,  # Direct Anthropic API
    FallbackLLMAdapter,   # Chain with fallback
)

# Gateway (recommended for production)
gateway = GatewayLLMAdapter(base_url="http://llm-gateway:8100")

# Fallback chain
fallback = FallbackLLMAdapter([
    GatewayLLMAdapter(base_url="http://llm-gateway:8100"),
    OpenAILLMAdapter(api_key="..."),
])

Django Integration

Add to INSTALLED_APPS:

INSTALLED_APPS = [
    ...
    "bfagent_llm",
]

Settings:

# LLM Gateway
LLM_GATEWAY_URL = "http://llm-gateway:8100"
LLM_GATEWAY_TIMEOUT = 120.0

# Direct API (optional)
OPENAI_API_KEY = "..."
ANTHROPIC_API_KEY = "..."

# Caching
REDIS_URL = "redis://localhost:6379/0"

License

MIT