vibeserver 0.3.0

Prompt-driven backend framework — add AI-powered natural language tool execution to any Python project

VibeServer

Prompt-driven backend framework. Replace REST endpoints with natural language.

pip install vibeserver

from vibeserver import VibeServer

vs = VibeServer()
@vs.tool("greet", description="Greet a user by name")
def greet(name: str) -> dict:
    return {"message": f"Hello, {name}!"}

vs.run()  # starts at http://localhost:8000

Then call it:

from vibeserver import VibeClient
async with VibeClient("http://localhost:8000") as client:
    result = await client.vibe("greet Alice")
    print(result.data)  # {"message": "Hello, Alice!"}

How It Works

Instead of POST /api/greet {"name": "Alice"}:

1. Send natural language: POST / with {"input": "greet Alice"}
2. LLM plans execution: {steps: [{action: "greet", params: {name: "Alice"}}]}
3. Tool executes and returns structured JSON
4. Repeat calls match the Intent Graph — no LLM needed

Quick Start

# Scaffold a new project
vibeserver init my-app
cd my-app

# Edit app.py to add your tools, then:
vibeserver serve

# Open http://localhost:8000 for the Web UI

SDK: Embedded Mode

from fastapi import FastAPI
from vibeserver import VibeServer

app = FastAPI()
vs = VibeServer()

@vs.tool("search_users", description="Search users by name")
def search_users(query: str, limit: int = 10) -> dict:
    results = db.query(f"SELECT * FROM users WHERE name LIKE '%{query}%' LIMIT {limit}")
    return {"users": results}

# Mount under existing FastAPI app
app.mount("/vibe", vs.asgi("my-app"))

Client: Remote Access

from vibeserver import VibeClient

async with VibeClient("http://vibe:8000", api_key="vs_xxx") as client:
    # Natural language execution
    result = await client.vibe("search users named Alice")
    
    # Streaming (SSE)
    async for event in client.stream("create a report for Q3"):
        print(event["type"], event)

    # Direct API
    tools = await client.list_tools()
    metrics = await client.metrics()

Built-in Tools (9 included)

Tool	Description
compute	Transform data (uppercase, lowercase, json_parse, sum, length)
create_record	Create a database record
search	Search records by type
update_record	Update a record
delete_record	Delete a record
get_user	Look up user by ID
auth	Verify API keys
list_users	List all users (admin)
stats	System statistics

Configuration

VIBESERVER_MODEL=openai/cmd/deepseek/deepseek-v4-pro
VIBESERVER_API_BASE=http://localhost:20128/v1
VIBESERVER_API_KEY=sk-xxx
VIBESERVER_DB_PATH=vibeserver.db
VIBESERVER_LOG_LEVEL=INFO

Or pass directly:

vs = VibeServer(
    model="openai/cmd/deepseek/deepseek-v4-pro",
    api_base="http://localhost:20128/v1",
    api_key="sk-xxx",
)

Docker

docker compose up

Architecture

POST / {"input": "create a note saying hello"}
  → Guardrails (prompt injection check)
  → Intent Graph: match cached plan?
      YES → ParamExtractor fills slots → execute
      NO  → LLM Planner → validate → new node → execute
  → Responder → {"status": "success", "data": {...}}

Full architecture docs: docs/ARCHITECTURE.md

License

MIT