reminix-runtime 0.0.8

Reminix Runtime - Serve AI agents as REST APIs with streaming support

reminix-runtime

Core runtime package for serving AI agents and tools via REST APIs. Provides the @agent, @chat_agent, and @tool decorators for building and serving AI agents.

Built on FastAPI with full async support.

Ready to go live? Deploy to Reminix Cloud for zero-config hosting, or self-host on your own infrastructure.

Installation

pip install reminix-runtime

Quick Start

from reminix_runtime import agent, chat_agent, serve, Message

# Create an agent for task-oriented operations
@agent
async def calculator(a: float, b: float) -> float:
    """Add two numbers."""
    return a + b

# Create a chat agent for conversational interactions
@chat_agent
async def assistant(messages: list[Message]) -> str:
    """A helpful assistant."""
    return f"You said: {messages[-1].content}"

# Serve the agents
serve(agents=[calculator, assistant], port=8080)

How It Works

The runtime creates a REST server with the following endpoints:

Endpoint	Method	Description
/health	GET	Health check
/info	GET	Runtime discovery (version, agents, tools)
/agents/{name}/execute	POST	Execute an agent
/tools/{name}/execute	POST	Execute a tool

Health Endpoint

curl http://localhost:8080/health

Returns {"status": "ok"} if the server is running.

Discovery Endpoint

curl http://localhost:8080/info

Returns runtime information, available agents, and tools:

{
  "runtime": {
    "name": "reminix-runtime",
    "version": "0.0.8",
    "language": "python",
    "framework": "fastapi"
  },
  "agents": [
    {
      "name": "calculator",
      "type": "agent",
      "description": "Add two numbers.",
      "parameters": {
        "type": "object",
        "properties": { "a": { "type": "number" }, "b": { "type": "number" } },
        "required": ["a", "b"]
      },
      "output": { "type": "number" },
      "requestKeys": ["a", "b"],
      "responseKeys": ["output"],
      "streaming": false
    },
    {
      "name": "assistant",
      "type": "chat_agent",
      "description": "A helpful assistant.",
      "parameters": {
        "type": "object",
        "properties": {
          "messages": {
            "type": "array",
            "items": { "type": "object", "properties": { "role": { "type": "string" }, "content": { "type": "string" } }, "required": ["role", "content"] }
          }
        },
        "required": ["messages"]
      },
      "output": { "type": "object" },
      "requestKeys": ["messages"],
      "responseKeys": ["message"],
      "streaming": false
    }
  ],
  "tools": [
    {
      "name": "get_weather",
      "type": "tool",
      "description": "Get current weather for a location",
      "parameters": { ... },
      "output": { ... }
    }
  ]
}

Agent Execute Endpoint

POST /agents/{name}/execute - Execute an agent.

Request keys are defined by the agent's parameters schema. For example, a calculator agent with parameters: { properties: { a, b } } expects a and b at the top level:

Task-oriented agent:

curl -X POST http://localhost:8080/agents/calculator/execute \
  -H "Content-Type: application/json" \
  -d '{"a": 5, "b": 3}'

Response:

{
  "output": 8.0
}

Chat agent:

Chat agents expect messages at the top level and return message:

curl -X POST http://localhost:8080/agents/assistant/execute \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Response:

{
  "message": {
    "role": "assistant",
    "content": "You said: Hello!"
  }
}

Tool Execute Endpoint

POST /tools/{name}/execute - Execute a standalone tool.

curl -X POST http://localhost:8080/tools/get_weather/execute \
  -H "Content-Type: application/json" \
  -d '{"location": "San Francisco"}'

Response:

{
  "output": { "temp": 72, "condition": "sunny" }
}

Agents

Agents handle requests via the /agents/{name}/execute endpoint.

Task-Oriented Agent

Use @agent for task-oriented agents that take structured input and return output:

from reminix_runtime import agent, serve

@agent
async def calculator(a: float, b: float) -> float:
    """Add two numbers."""
    return a + b

@agent(name="text-processor", description="Process text in various ways")
async def process_text(text: str, operation: str = "uppercase") -> str:
    """Process text with the specified operation."""
    if operation == "uppercase":
        return text.upper()
    elif operation == "lowercase":
        return text.lower()
    return text

serve(agents=[calculator, process_text], port=8080)

The decorator automatically extracts:

• name from the function name (or use name= to override)
• description from the docstring (or use description= to override)
• parameters from type hints and defaults
• output from the return type hint

Chat Agent

Use @chat_agent for conversational agents that handle message history:

from reminix_runtime import chat_agent, serve, Message

@chat_agent
async def assistant(messages: list[Message]) -> str:
    """A helpful assistant."""
    last_msg = messages[-1].content
    return f"You said: {last_msg}"

# With context support
@chat_agent
async def contextual_bot(messages: list[Message], context: dict | None = None) -> str:
    """Bot with context awareness."""
    user_id = context.get("user_id") if context else "unknown"
    return f"Hello user {user_id}!"

serve(agents=[assistant, contextual_bot], port=8080)

Streaming

Both decorators support streaming via async generators. When you use yield instead of return, the agent automatically supports streaming:

from reminix_runtime import agent, chat_agent, serve, Message

# Streaming task agent
@agent
async def streamer(text: str):
    """Stream text word by word."""
    for word in text.split():
        yield word + " "

# Streaming chat agent
@chat_agent
async def streaming_assistant(messages: list[Message]):
    """Stream responses token by token."""
    response = f"You said: {messages[-1].content}"
    for char in response:
        yield char

serve(agents=[streamer, streaming_assistant], port=8080)

For streaming agents:

• stream: true in the request → chunks are sent via SSE
• stream: false in the request → chunks are collected and returned as a single response

Tools

Tools are standalone functions served via /tools/{name}/execute. They're useful for exposing utility functions, external API integrations, or any reusable logic.

Creating Tools

Use the @tool decorator to create tools:

from reminix_runtime import tool, serve

@tool
async def get_weather(location: str, units: str = "celsius") -> dict:
    """Get current weather for a location."""
    # Call weather API...
    return {"temp": 72, "condition": "sunny", "location": location}

@tool
def calculate(expression: str) -> dict:
    """Evaluate a math expression."""
    return {"result": eval(expression)}  # Note: use a safe evaluator in production

serve(tools=[get_weather, calculate], port=8080)

The decorator automatically extracts:

• name from the function name
• description from the docstring
• parameters from type hints and defaults
• output from the return type hint

Custom Tool Configuration

You can customize the tool name and description:

@tool(name="weather_lookup", description="Look up weather for any city")
async def get_weather(location: str) -> dict:
    return {"temp": 72, "condition": "sunny"}

Serving Agents and Tools Together

You can serve both agents and tools from the same runtime:

from reminix_runtime import agent, tool, serve

@agent
async def summarizer(text: str) -> str:
    """Summarize text."""
    return text[:100] + "..."

@tool
def calculate(expression: str) -> dict:
    """Perform basic math operations."""
    return {"result": eval(expression)}

serve(agents=[summarizer], tools=[calculate], port=8080)

Framework Adapters

Instead of creating custom agents, use our pre-built adapters for popular frameworks:

Package	Framework
reminix-langchain	LangChain
reminix-langgraph	LangGraph
reminix-openai	OpenAI
reminix-anthropic	Anthropic
reminix-llamaindex	LlamaIndex

API Reference

serve(agents, tools, port, host)

Start the runtime server.

Parameter	Type	Default	Description
agents	list[Agent]	[]	List of agents to serve
tools	list[Tool]	[]	List of tools to serve
port	int	8080	Port to listen on. Falls back to PORT environment variable if not provided.
host	str	"0.0.0.0"	Host to bind to (all interfaces). Can be overridden via HOST env var.

At least one agent or tool is required.

create_app(agents, tools)

Create a FastAPI app without starting the server. Useful for testing or custom deployment.

from reminix_runtime import create_app

app = create_app(agents=[my_agent], tools=[my_tool])
# Use with uvicorn, gunicorn, etc.

@agent

Decorator to create a task-oriented agent from a function.

from reminix_runtime import agent

@agent
async def my_agent(param: str, count: int = 5) -> str:
    """Agent description from docstring."""
    return param * count

# With custom name/description
@agent(name="custom_name", description="Custom description")
async def another_agent(x: int) -> int:
    return x * 2

# Streaming agent
@agent
async def streaming_agent(text: str):
    for word in text.split():
        yield word + " "

@chat_agent

Decorator to create a chat agent from a function.

from reminix_runtime import chat_agent, Message

@chat_agent
async def my_chat_agent(messages: list[Message]) -> str:
    """Chat agent description."""
    return f"You said: {messages[-1].content}"

# With context
@chat_agent
async def contextual_agent(messages: list[Message], context: dict | None = None) -> str:
    user_id = context.get("user_id") if context else None
    return f"Hello user {user_id}!"

# Streaming chat agent
@chat_agent
async def streaming_chat(messages: list[Message]):
    for token in ["Hello", " ", "world!"]:
        yield token

@tool

Decorator to create a tool from a function.

from reminix_runtime import tool

@tool
async def my_tool(param: str, optional_param: int = 10) -> dict:
    """Tool description from docstring."""
    return {"result": param, "value": optional_param}

# With custom name/description
@tool(name="custom_name", description="Custom description")
def another_tool(x: int) -> int:
    return x * 2

Request/Response Types

# Request: top-level keys based on agent's requestKeys (derived from parameters)
# For a calculator agent with parameters { a: float, b: float }:
# {
#   "a": 5,                         # Top-level key from parameters
#   "b": 3,                         # Top-level key from parameters  
#   "stream": false,                # Whether to stream the response
#   "context": { ... }              # Optional metadata
# }

# For a chat agent:
# {
#   "messages": [...],              # Top-level key (requestKeys: ['messages'])
#   "stream": false,
#   "context": { ... }
# }

# Response: keys based on agent's responseKeys
# Regular agent (responseKeys: ['output']):
# { "output": ... }

# Chat agent (responseKeys: ['message']):
# { "message": { "role": "assistant", "content": "..." } }

Advanced

Agent Class

For more control, you can use the Agent class directly:

from reminix_runtime import Agent, ExecuteRequest, ExecuteResponse, serve

agent = Agent("my-agent", metadata={"version": "1.0"})

@agent.on_execute
async def handle_execute(request: ExecuteRequest) -> ExecuteResponse:
    return ExecuteResponse(output="Hello!")

# Optional: streaming handler
@agent.on_execute_stream
async def handle_execute_stream(request: ExecuteRequest):
    yield "Hello"
    yield " world!"

serve(agents=[agent], port=8080)

Tool Class

For programmatic tool creation:

from reminix_runtime import Tool, ToolSchema, ToolExecuteRequest, ToolExecuteResponse, serve

async def execute_handler(request: ToolExecuteRequest) -> ToolExecuteResponse:
    location = request.input.get("location", "unknown")
    return ToolExecuteResponse(output={"temp": 72, "location": location})

my_tool = Tool(
    execute_handler,
    name="get_weather",
    description="Get weather for a location",
)

serve(tools=[my_tool], port=8080)

AgentAdapter

For building framework integrations. See the framework adapter packages for examples.

from reminix_runtime import AgentAdapter, ExecuteRequest, ExecuteResponse

class MyFrameworkAdapter(AgentAdapter):
    adapter_name = "my-framework"

    def __init__(self, client, name: str = "my-framework"):
        self._client = client
        self._name = name

    @property
    def name(self) -> str:
        return self._name

    async def execute(self, request: ExecuteRequest) -> ExecuteResponse:
        result = await self._client.run(request.input)
        return ExecuteResponse(output=result)

Serverless Deployment

Use to_asgi() for serverless deployments:

# AWS Lambda with Mangum
from mangum import Mangum
from reminix_runtime import agent, ExecuteResponse

@agent
async def my_agent(task: str) -> str:
    return f"Completed: {task}"

handler = Mangum(my_agent.to_asgi())

Deployment

Ready to go live?

• Deploy to Reminix Cloud - Zero-config cloud hosting
• Self-host - Run on your own infrastructure

Links

• GitHub Repository
• Documentation

License

Apache-2.0