01Install
LM Studio runs models locally with an OpenAI-compatible server. No cloud API key required.
Install
bashpip install "mtpx[lmstudio]"This installs the openai SDK (used for the OpenAI-compatible API).
bashpip install openai02Setup
Setup
- Download LM Studio from lmstudio.ai
- Download a tool-capable model (e.g., Qwen 3, Llama 3.2)
- Start the local server (default:
http://127.0.0.1:1234/v1) - Load a model in the LM Studio UI
No API key is required for local usage. A dummy "lm-studio" key is used by default. If your LM Studio instance requires auth, set LMSTUDIO_API_KEY in your .env file.
03Quick Start
Quick Start
pythonfrom mtp import Agent
from mtp.providers import LMStudio
# No API key needed for local LM Studio
provider = LMStudio(model="qwen3-4b-thinking-2507")
tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools)
reply = agent.run_loop("What is 25 * 4 + 10?")
print(reply)04Parameters
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
model | str | "qwen3" | Model name as loaded in LM Studio |
base_url | str | "http://127.0.0.1:1234/v1" | LM Studio server URL |
api_key | str | None | None | API key (falls back to LMSTUDIO_API_KEY env var, then "lm-studio") |
temperature | float | 0.0 | Sampling temperature |
tool_choice | str | dict | "auto" | Tool selection strategy |
parallel_tool_calls | bool | True | Allow parallel tool calls |
client | Any | None | None | Pre-configured openai.OpenAI client instance |
05Capabilities
Capabilities
| Capability | Value |
|---|---|
| Tool calling | Yes (if model supports it) |
| Parallel tool calls | Yes (configurable) |
| Input modalities | text, image |
| Streaming | Yes (both stream_next_action and finalize_stream) |
| Usage metrics | Rich |
| Reasoning metadata | No |
| Native async | No (uses thread fallback) |
06Streaming
Streaming
LM Studio supports both planning and finalization streaming with reasoning chunk support:
pythonfor event in agent.run_loop_events(
"Calculate 25 * 4",
max_rounds=2,
stream_final=True,
):
if isinstance(event, dict):
if event.get("type") == "reasoning_chunk":
print("[thinking]", event["chunk"])
elif event.get("type") == "text_chunk":
print(event["chunk"], end="", flush=True)07Full Example
Full Example
pythonfrom mtp import Agent
from mtp.providers import LMStudio
provider = LMStudio(
model="qwen3-4b-thinking-2507",
base_url="http://127.0.0.1:1234/v1",
temperature=0.0,
tool_choice="auto",
parallel_tool_calls=True,
)
tools = Agent.ToolRegistry()
agent = Agent(provider=provider, tools=tools, debug_mode=True)
reply = agent.run_loop(
"Calculate (25 * 4) + 10",
max_rounds=3,
)
print(reply)08Notes
Notes
- LM Studio uses the OpenAI SDK under the hood, pointed at the local server.
- Tool calling support depends on the loaded model. Use Qwen 3, Llama 3.2, or other tool-capable models.
- The
parallel_tool_callsparameter may be ignored by some model/SDK versions (handled gracefully with fallback).
09Source
Source
src/mtp/providers/lmstudio_provider.py