Skip to content

RELEASE-NOTES-v0.1.0.md

Latest commit

 

History

History
144 lines (101 loc) · 3.99 KB

RELEASE-NOTES-v0.1.0.md

File metadata and controls

144 lines (101 loc) · 3.99 KB
Raw

Agent Observability Kit v0.1.0

Release Date: February 4, 2026

🎉 Initial Release

The first public release of the Agent Observability Kit - framework-agnostic visual debugging for AI agents.

🚀 What's Included

Core Tracing SDK:

  • ✅ Universal @observe decorator for any Python function
  • ✅ Context manager API (with trace())
  • ✅ LLM call tracking (model, tokens, cost, latency)
  • ✅ Error capture with full stack traces
  • ✅ JSON-based storage (local-first, no cloud required)

Framework Integrations:

  • ✅ LangChain: Drop-in callback handler for automatic tracing
  • ✅ OpenClaw: Native integration for OpenClaw agents
  • 🚧 CrewAI and AutoGen (coming in v0.2.0)

Web Visualization UI:

  • ✅ Real-time dashboard with trace metrics
  • ✅ Interactive execution graph visualization
  • ✅ Step-level inspection (click any node to see details)
  • ✅ LLM call viewer (prompts, responses, tokens, cost)
  • ✅ Error highlighting and debugging

Documentation:

  • ✅ Comprehensive README with examples
  • ✅ Quick Start guide (5-minute setup)
  • ✅ API documentation
  • ✅ Working examples (basic + LangChain)

🎯 Key Features

Framework-Agnostic: Unlike LangGraph Studio (LangGraph-only) or LangSmith (LangChain-focused), this toolkit works with ANY Python-based agent framework.

Visual Debugging: See your agent's execution flow as an interactive graph. Click any step to inspect inputs, outputs, LLM calls, and errors.

Local-First: All traces stored locally on your machine. No cloud dependencies, no API keys, no vendor lock-in.

Production-Ready: Low overhead (<1% latency), async data collection, and configurable retention.

📊 Use Cases

  1. Debug multi-agent systems - See which agent failed and why
  2. Optimize performance - Identify slow steps and bottlenecks
  3. Track LLM costs - See token usage and costs per operation
  4. Root cause analysis - Inspect full error traces with context

🔧 Quick Start

# Install
pip install agent-observability-kit

# Decorate your functions
from agent_observability import observe, init_tracer
from agent_observability.span import SpanType

tracer = init_tracer(agent_id="my-agent")

@observe(span_type=SpanType.AGENT_DECISION)
def my_agent_function(input):
    # Your code here
    return result

# Start web UI
python -m agent_observability.server

# View at http://localhost:5000

🐛 Known Issues

  • Server requires manual start (no CLI entry point yet)
  • No real-time streaming of traces (5-second polling)
  • Limited filtering/search in UI
  • No production monitoring features yet

🗺️ Roadmap

v0.2.0 (4 weeks):

  • CrewAI and AutoGen integrations
  • Real-time trace streaming (WebSocket)
  • Advanced filtering and search
  • Trace comparison tool

v0.3.0 (8 weeks):

  • Production monitoring dashboard
  • Cost alerts and budgets
  • Quality metrics (accuracy, latency)
  • Anomaly detection

v1.0.0 (12 weeks):

  • Self-hosted deployment (Docker, K8s)
  • Multi-tenancy and RBAC
  • PII redaction
  • Enterprise features

📦 Installation

pip install agent-observability-kit

Or from source:

git clone https://github.com/reflectt/agent-observability-kit.git
cd agent-observability-kit
pip install -e .

🤝 Contributing

We welcome contributions! Priority areas:

  • Framework integrations (CrewAI, AutoGen, etc.)
  • UI improvements (filtering, search, real-time updates)
  • Performance optimizations
  • Documentation and examples

See CONTRIBUTING.md for guidelines.

📄 License

Apache 2.0 - See LICENSE for details.

🙏 Acknowledgments

Inspired by:

  • LangGraph Studio - Best-in-class visual debugging
  • LangSmith - Production observability for LLMs
  • OpenTelemetry - Distributed tracing standards

Built by the OpenClaw team and Reflectt.

Questions or feedback? Open an issue or join our Discord

Star the repo if you find this useful! ⭐