PandaProbe

PandaProbe: The Complete Open Source Agent Engineering Platform

In the rapidly evolving landscape of artificial intelligence, building reliable agents requires more than just code; it requires deep visibility and rigorous evaluation. PandaProbe is a premier open-source agent engineering platform designed to provide developers with the necessary infrastructure to debug, monitor, and improve AI agents. Built by Chirpz AI, the PandaProbe platform is licensed under Apache 2.0, offering a self-hostable solution that prevents vendor lock-in while remaining built for scale.

Whether you are building your first agent or scaling a complex production system, PandaProbe offers a unified platform for the full agent development lifecycle—from the initial run to continuous improvement. By integrating PandaProbe into your stack, you gain access to sophisticated traces, evals, metrics, and live monitoring capabilities.

What's PandaProbe?

PandaProbe is an open-source agent engineering platform tailored specifically for the needs of AI developers. At its core, PandaProbe serves as an observability and optimization layer for AI agents. It allows developers to capture every step of an agent's execution, providing a clear window into how LLMs, tools, and chains interact in real-time.

As a product of Chirpz AI, PandaProbe emphasizes flexibility and scalability. It is available as both a managed PandaProbe Cloud service and a self-hosted Open Source version. The platform is designed to work seamlessly with any stack, featuring a robust Python SDK and plug-and-play integrations with leading agent frameworks and LLM providers.

Core Features of PandaProbe

To bridge the gap between a prototype and a production-ready AI agent, PandaProbe provides a suite of powerful features focused on observability and performance.

1. Advanced Tracing

Tracing is the backbone of the PandaProbe experience. With a simple instrumentation call, developers can capture every step of an agent's execution sequence. This feature allows you to:

• Capture Every Step: Use a single instrument() call to trace the full agent run automatically.
• Universal Compatibility: Plug-and-play with every top agent framework and work seamlessly with any LLM provider.
• Detailed Visibility: Instantly see every span, including chains, agents, LLMs, tool calls, and more.
• Metadata Tracking: Monitor model types, parameters, token usage, and key metadata to understand the cost and performance of every interaction.

2. Evaluations (Evals) & Metrics

Improving an agent requires data-driven decisions. PandaProbe enables developers to run evaluations and track metrics to ensure the agent's output meets quality standards. By analyzing trace data, PandaProbe helps identify bottlenecks in the agent's logic or failures in tool usage.

3. Continuous Monitoring

Once an agent is in production, PandaProbe provides live monitoring to ensure ongoing reliability. It tracks crucial metrics such as Time to First Token (TTFT) and total token usage, allowing teams to maintain high performance and manage operational costs effectively.

Use Case Scenarios

PandaProbe is versatile enough to support a variety of development and production use cases:

• Debugging Complex Agents: When an agent fails to complete a task, PandaProbe's tracing allows you to pinpoint exactly which tool or LLM hop caused the error.
• Performance Optimization: By tracking metrics like token usage and TTFT, developers can optimize their agents for speed and cost-efficiency.
• Production Scaling: For large-scale deployments, PandaProbe’s self-hostable architecture and high rate limits in the Startup and Enterprise plans ensure the platform grows with your user base.
• Framework Integration: Developers using LangChain, LangGraph, CrewAI, or Google ADK can integrate PandaProbe to get immediate visibility without rewriting their core logic.

How to Use PandaProbe

Getting started with PandaProbe is straightforward thanks to its developer-centric Python SDK. The platform is designed to be integrated with minimal code changes.

Initializing the SDK

To begin tracing your agents, you first need to set up the adapter. Below is an example of how to use the Google ADK Adapter within your Python environment:

from pandaprobe.integrations.google_adk import GoogleADKAdapter

# Call once at startup — before creating any agents
adapter = GoogleADKAdapter(
    session_id="session-abc",
    user_id="user-123",
    tags=["production"],
)

adapter.instrument()

# All ADK runners are now fully traced
# — tool calls, LLM hops, token usage, TTFT

By calling adapter.instrument(), you enable full visibility across all ADK runners, capturing everything from tool calls to LLM responses automatically.

Supported Integrations

PandaProbe is built to work with the tools you already use. Its Python SDK features seamless integrations with:

• Frameworks: LangGraph, LangChain, CrewAI, Google ADK, Claude Agent SDK, and OpenAI Agents SDK.
• LLM Providers: OpenAI, Gemini, and Anthropic.

Pricing and Plans

PandaProbe offers a range of pricing tiers to suit everyone from individual hobbyists to large enterprises.

PandaProbe Cloud

• Hobby ($0/forever): Ideal for getting started. Includes 100 base trace ingestions/mo, 100 trace eval runs/mo, 10 session eval runs/mo, human annotation, 1 seat, and community support.
• Pro ($29/month): For developers and small teams. Includes 5k base traces/mo (then pay-as-you-go), 5k trace eval runs/mo, 100 session eval runs/mo, 2 seats, and email support.
• Startup ($299/mo): For scaling projects. Includes 50k base traces/mo, 50k trace eval runs/mo, 1k session eval runs/mo, 10 seats, high rate limits, a private Slack channel, and data retention management.
• Enterprise (Custom): For large organizations needing hybrid/self-hosted options, custom SSO, dedicated engineering support, and unlimited seats.

Open Source

• OSS (Free): Self-host all core PandaProbe features for free without any limitations. This includes the Apache 2.0 license, all core platform features, and the same scalability found in the Cloud version.

FAQ

What is PandaProbe? PandaProbe is an open-source agent engineering platform by Chirpz AI, designed for tracing, evaluating, and monitoring AI agents.

What does PandaProbe help me with? It helps you debug agent runs, track performance metrics (like token usage and TTFT), and continuously improve agent quality through evaluations.

Can I use just tracing without the other features? Yes, the platform is flexible. You can use the instrument() call specifically for tracing to gain visibility into your agent spans, chains, and tool calls.

What deployment options exist? You can choose PandaProbe Cloud, where the hosting is managed for you, or you can choose to Self-host the platform on your own infrastructure.

Is self-hosting actually free? Yes, the Open Source version is free under the Apache 2.0 license and includes all core features without limitations.

What frameworks are supported? PandaProbe supports major frameworks including LangChain, LangGraph, CrewAI, Google ADK, and SDKs from OpenAI, Claude, and Anthropic.

What's the latency impact? PandaProbe is built for scale and designed to have minimal impact on your agent's performance, while providing real-time monitoring of metrics like TTFT.

How do I get started? You can start with the Hobby plan for free or deploy the Open Source version. Simply install the Python SDK and use the instrumentation calls to begin tracing.

How does pricing work? PandaProbe uses a tiered model based on the number of trace ingestions and evaluation runs, with a free hobby tier and a fully free self-hosted open-source option.