Agent Framework Integrations

Agent Framework Integrations LOCAL MANAGED

Privane integrates seamlessly with modern multi-agent orchestration frameworks. Because our background CLI server exposes a standard, highly efficient OpenAI-compatible REST API, you can utilize Privane as a drop-in local execution engine for any major agentic library.

By offloading the complex, multi-step local reasoning loops to your own hardware, you drastically reduce cloud API costs while maintaining absolute privacy for local documents and files.

Supported Ecosystem Frameworks

You can instantly bind Privane to:

  • LangChain / LangGraph: Build robust, stateful graph-based agents.
  • AutoGen (Microsoft): Orchestrate multi-agent conversational structures locally.
  • CrewAI: Dispatched role-playing teams of agents working in parallel.
  • OpenHands (formerly OpenDevin): Drive autonomous software engineering loops.
  • Semantic Kernel (Microsoft): Blend local models with enterprise plugins.
  • Google Agent Development Kit: Coordinate fast local workflows powered by Gemma.

The Value: Local Reasoning Loops

Multi-step agent workflows are famous for executing massive loops of sub-queries (planning, self-critique, tool selection, and parsing). Dispatched over proprietary cloud endpoints like GPT-4 or Claude, a single complex task can easily burn through $2.00 to $10.00 in API costs.

With Privane:

  1. The Reasoning Loop executes entirely on-device (free and unlimited).
  2. The Execution Gate triggers cloud actions only when a real-world task needs to occur, metered safely by our managed connector runtime.

Code Examples

1. LangChain (TypeScript)

import { ChatOpenAI } from "@langchain/openai";
 
// Bind LangChain's standard interface to your local execution engine
const model = new ChatOpenAI({
  configuration: {
    baseURL: "http://localhost:8080/v1",
  },
  apiKey: "privane-local",
  modelName: "gemma-2b-instruct",
  temperature: 0.2
});
 
const response = await model.invoke("Plan a 3-step refactoring workflow.");
console.log(response.content);

2. Google Agent Kit (Python)

from google_agent_kit import Agent, Tool
from google_agent_kit.llms import OpenAIChatModel
 
# Route Google's Agent framework to local Gemma weights
local_gemma = OpenAIChatModel(
    base_url="http://localhost:8080/v1",
    api_key="privane-local", 
    model="gemma-2b-instruct"
)
 
def search_local_workspace(query: str) -> str:
    """Scrapes the local project folder for relevant code files."""
    return f"Indexed 5 files matching '{query}'"
 
agent = Agent(
    llm=local_gemma,
    tools=[Tool(search_local_workspace)],
    system_prompt="You are a localized software assistant. Access local files safely."
)
 
response = agent.run("Find our recent package dependencies.")
print(response)

3. Microsoft AutoGen (Python)

import autogen
 
# Configure AutoGen to route agent conversations through Privane serve
config_list = [{
    "model": "gemma-2b-instruct",
    "api_key": "privane-local",
    "base_url": "http://localhost:8080/v1",
    "price": [0, 0] # Zero API Cost for local reasoning loops!
}]
 
assistant = autogen.AssistantAgent(
    name="coder",
    llm_config={"config_list": config_list}
)
 
user_proxy = autogen.UserProxyAgent(
    name="user",
    code_execution_config={"work_dir": "workspace"}
)
 
user_proxy.initiate_chat(assistant, message="Analyze our local database schema.")
Pricing & LimitsMobile SDKs (Edge)