dspy.context¶

Override DSPy settings inside one with block.

with dspy.context(**kwargs):
    ...  # overrides active here
# original settings restored

Use dspy.context(...) when you need a different LM, adapter, or flag for one part of your program without changing the process-wide defaults from dspy.configure. The block inherits every current setting, overrides only the keys you pass, and restores the originals when it exits.

dspy.context(...) accepts the same settings as dspy.configure.

Examples¶

Use a different LM for one block¶

import dspy

dspy.configure(lm=dspy.LM("openai/gpt-5-mini"))
qa = dspy.Predict("question -> answer")

result = qa(question="What is the capital of France?")
print("default:", result.answer)

with dspy.context(lm=dspy.LM("anthropic/claude-sonnet-4-6")):
    result = qa(question="What is the capital of France?")
    print("temporary:", result.answer)

print("restored:", dspy.settings.lm.model)  # openai/gpt-5-mini

Use a different adapter for one block¶

import dspy

dspy.configure(lm=dspy.LM("gemini/gemini-3-flash-preview"))
qa = dspy.Predict("question -> answer")

with dspy.context(adapter=dspy.JSONAdapter()):
    result = qa(question="What is the capital of France?")
    print(result.answer)

Enable async tool conversion temporarily¶

import asyncio
import dspy

async def async_tool(x: int) -> int:
    await asyncio.sleep(0.1)
    return x * 2

tool = dspy.Tool(async_tool)

with dspy.context(allow_tool_async_sync_conversion=True):
    print(tool(x=5))

Nested blocks¶

Inner blocks override outer ones. Each restores cleanly on exit:

import dspy

dspy.configure(lm=dspy.LM("openai/gpt-5-mini"))

with dspy.context(lm=dspy.LM("anthropic/claude-sonnet-4-6")):
    print(dspy.settings.lm.model)        # anthropic/claude-sonnet-4-6
    with dspy.context(track_usage=True):
        print(dspy.settings.lm.model)    # anthropic/claude-sonnet-4-6 (inherited)
        print(dspy.settings.track_usage) # True
    print(dspy.settings.track_usage)     # False (restored)
print(dspy.settings.lm.model)           # openai/gpt-5-mini (restored)

Thread safety¶

Unlike dspy.configure, you can call dspy.context(...) from any thread or async task. This makes it the right tool for overrides inside dspy.Parallel, asyncio.gather, or any concurrent code.

Settings inside a dspy.context(...) block do not leak to other threads or tasks.