opperai 2.0.0

Python SDK for the Opper Task API

Opper Python SDK

Python client for the Opper API.

Install

pip install opperai

Quick Start

from opperai import Opper

opper = Opper()  # uses OPPER_API_KEY env var

result = opper.call("summarize", input={"text": "Long article..."})
print(result.data)

# Stream a function
for chunk in opper.stream("summarize", input={"text": "Long article..."}):
    if chunk.type == "content":
        print(chunk.delta, end="")
    if chunk.type == "complete":
        print(chunk.data)

Schema Support

Pass Pydantic models, dataclasses, TypedDicts, or raw JSON Schema dicts for input_schema and output_schema — the SDK resolves them to JSON Schema automatically.

from pydantic import BaseModel

class Summary(BaseModel):
    summary: str
    entities: list[str]

result = opper.call(
    "extract",
    input={"text": "Marie Curie was a physicist in Paris."},
    output_schema=Summary,
)
result.data.summary   # str — typed!
result.data.entities  # list[str]

Dataclasses, TypedDicts, and plain dicts also work. See 01a_using_schemas.py and 01b_using_other_schemas.py.

Observability

Use trace() as a decorator or context manager to group calls under a single trace span. Nesting works naturally.

@opper.trace("my-pipeline")
def run():
    a = opper.call("step-1", input="hello")
    b = opper.call("step-2", input=a.data)

# or as a context manager
with opper.trace("my-pipeline") as span:
    opper.call("step-1", input="hello")

Agent SDK

Build AI agents with tool use, streaming, multi-agent composition, and MCP integration.

from opperai import Agent, tool

@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"Sunny, 22°C in {city}"

agent = Agent(
    name="weather-assistant",
    instructions="You are a helpful weather assistant.",
    tools=[get_weather],
)

# Run — get the final result
result = await agent.run("What's the weather in Paris?")
print(result.output)
print(result.meta.usage)  # token usage across all iterations

# Stream — observe events as the agent works
stream = agent.stream("What's the weather in Paris?")
async for event in stream:
    if event.type == "text_delta":
        print(event.text, end="", flush=True)
    if event.type == "tool_start":
        print(f"\nCalling {event.name}...")
result = await stream.result()

Structured Output

from pydantic import BaseModel

class Sentiment(BaseModel):
    label: str
    score: float

agent = Agent(
    name="analyzer",
    instructions="Analyze the sentiment of the input.",
    output_schema=Sentiment,
)

result = await agent.run("I love this product!")
result.output.label  # str — typed via Pydantic
result.output.score  # float

Multi-Agent Composition

researcher = Agent(name="researcher", instructions="...", tools=[web_search])
writer = Agent(
    name="writer",
    instructions="Write clear reports using research.",
    tools=[researcher.as_tool(name="research", description="Research a topic")],
)

result = await writer.run("Write a report on AI agents")

MCP Integration

from opperai.agent.mcp import mcp, MCPStdioConfig

agent = Agent(
    name="file-assistant",
    instructions="Help users manage files.",
    tools=[mcp(MCPStdioConfig(name="fs", command="uvx", args=["mcp-server-filesystem", "/tmp"]))],
)

Conversation (Multi-Turn)

conversation = agent.conversation()
r1 = await conversation.send("My name is Alice")
r2 = await conversation.send("What is my name?")
# r2.output → "Your name is Alice"

Examples

#	Example	What it shows
00	First call	Simplest possible call
01a	Pydantic schemas	Type-safe output with Pydantic
01b	Other schemas	Dataclass, TypedDict, raw dict
02	Streaming	Stream deltas + complete event
03a	Tools (call)	Tool definitions with call()
03b	Tools (stream)	Tool call chunks in streaming
04a	Generate image	Image generation
04b	Describe image	Vision / image description
04c	Edit image	Image editing
05	Audio	Text-to-speech + speech-to-text
06	Video	Video generation
07	Embeddings	Vector embeddings + similarity
08	Function mgmt	List, get, revisions, delete
09	Observability	Tracing with decorator + context manager
09b	Manual tracing	Manual span creation
09c	Traces	List, get, and inspect traces
10	Models	List available models
11	Realtime	Mint a ticket for browser-direct voice WebSocket
12	Knowledge base	Semantic search with knowledge bases
13	Web tools	Web search and URL fetch (beta)

Run a single example:

export OPPER_API_KEY="your-key"
uv run python examples/getting-started/00_your_first_call.py

Run all examples:

uv run python examples/run_all.py

Configuration

Parameter	Default	Env Var
api_key	—	OPPER_API_KEY
base_url	https://api.opper.ai	OPPER_BASE_URL
headers	{}	—

Error Handling

from opperai import ApiError

try:
    opper.call("my-fn", input="hello")
except ApiError as e:
    print(e.status, e.body)

Async Support

All methods have _async variants:

result = await opper.call_async("summarize", input={"text": "..."})

async for chunk in opper.stream_async("summarize", input={"text": "..."}):
    print(chunk.delta, end="")

Requirements

• Python 3.10+
• Optional: pip install opperai[pydantic] for Pydantic schema support

License

MIT