wildedge-sdk 0.3.0

Monitor local and remote ML inference from Python. Track latency, token usage, errors, model metadata, and local hardware context across llama.cpp, MLX, Transformers, OpenAI, and more.

Instrument supported libraries with one command, or add tracking in your code. Prompts and outputs aren't uploaded unless you explicitly attach them.

Get started · Explore the demo · Docs · Website

Model performance	Workflow trace	SQL editor

Illustrative sample data across WildEdge SDKs. Click a thumbnail for the full-size image, or explore the read-only demo.

More screenshots: events and hardware

Events	Hardware

Install

uv add wildedge-sdk

To send events, sign up at app.wildedge.dev, create a project and copy its DSN. Without a DSN everything is a silent no-op, so dev and CI need no configuration.

CLI

Monitor an existing application by adding wildedge run to its start command:

WILDEDGE_DSN="https://<secret>@ingest.wildedge.dev/<key>" \
uv run wildedge run --integrations llama_cpp -- python app.py

This assumes your application already uses llama-cpp-python. The llama_cpp integration instruments that library; replace it with your integration. No SDK calls are required in your application code.

See the deployment guide for startup options, environment validation, and production configuration.

SDK

To try tracking in your code, run one local text-generation model on CPU. Install the model's dependencies:

uv add transformers torch

Save this as app.py. The first run downloads the model from Hugging Face; no model-provider API key is needed.

import wildedge
from transformers import pipeline

client = wildedge.init(integrations=["transformers"])
try:
    model = pipeline(
        "text-generation",
        model="HuggingFaceTB/SmolLM2-360M-Instruct",
        device=-1,  # CPU
    )
    result = model(
        "Explain how rain forms in one sentence.",
        max_new_tokens=32,
        do_sample=False,
        return_full_text=False,
    )
    print(result[0]["generated_text"])
finally:
    client.close(timeout=10)

Set the DSN from your WildEdge project and run it:

export WILDEDGE_DSN="https://<secret>@ingest.wildedge.dev/<key>"
uv run python app.py

Open your project at app.wildedge.dev to inspect the model load and inference events, including duration and model metadata. The generated text is printed locally; it isn't uploaded.

Already have an application? Initialize WildEdge before creating your model or API client. See examples for each integration, traces and spans, and the deployment guide for lifecycle details.

import wildedge
from openai import OpenAI
from transformers import pipeline

client = wildedge.init(integrations=["transformers", "openai"])
try:
    local = pipeline(
        "text-generation",
        model="HuggingFaceTB/SmolLM2-360M-Instruct",
        device=-1,
    )
    remote = OpenAI()
    prompt = "Explain how rain forms in one sentence."

    with wildedge.trace(run_id="run-1"):
        with wildedge.span(kind="agent_step", name="draft"):
            draft = local(
                prompt, max_new_tokens=32, return_full_text=False
            )[0]["generated_text"]
        with wildedge.span(kind="agent_step", name="refine"):
            response = remote.chat.completions.create(
                model="gpt-4o-mini",
                messages=[{"role": "user", "content": draft}],
            )
            print(response.choices[0].message.content)
finally:
    client.close(timeout=10)

Supported integrations

Local inference: llama.cpp, Transformers, MLX-LM, ONNX Runtime, PyTorch, TensorFlow, Keras, Ultralytics, timm, faster-whisper, and MLX Whisper.

API clients: OpenAI, Anthropic, and OpenAI-compatible providers.

Most integrations track supported entry points automatically after activation. PyTorch and standalone Keras models require explicit registration.

For unsupported frameworks, see Manual tracking.

Configuration

For advanced options (batching, queue tuning, dead-letter storage, attachments), see Configuration.

Security & Privacy

Automatic integrations send inference metrics, model metadata, and device/hardware information, including a device identifier, OS information, locale, and timezone. They do not attach raw inputs or outputs. Raw payload uploads require explicitly supplied attachments and attachments_enabled=True.

Report security and privacy issues to: support@wildedge.dev

Links

• Deployment guide
• Manual tracking
• LLM API tracking
• Compatibility Matrix
• Streaming lifecycle
• Changelog
• License