Agent Observability Studio
Real-time monitoring and debugging platform for multi-agent AI systems—Chrome DevTools for AI agents.
What is this?
Agent Observability Studio provides deep inspection tools for multi-agent AI systems, complementing existing infrastructure dashboards. It captures live interaction traces, analyzes token costs, detects performance bottlenecks, and helps debug agent decision trees and failures in production environments.
Features
- Live Interaction Tracing — Real-time visibility into agent conversations and inter-agent communication
- Token Cost Analysis — Granular per-agent, per-interaction token accounting with cost attribution
- Performance Bottleneck Detection — Automatic identification of slow agents, network delays, and processing inefficiencies
- Decision Tree Inspection — Visualize agent reasoning paths and fallback logic execution
- Failure Analysis — Capture and replay failed interactions with full context
- WebSocket Streaming — Push-based updates for low-latency monitoring
- RESTful API — Programmatic access to traces, metrics, and historical data
- CLI Tools — Command-line utilities for local debugging and integration
Quick Start
Installation
pip install agent-observability-studio
Basic Setup
from agent_observability_studio import ObservabilityClient
# Initialize the client
client = ObservabilityClient(
api_url="http://localhost:8000",
api_key="your-api-key"
)
# Start monitoring an agent interaction
trace = client.start_trace(
agent_id="my-agent",
session_id="session-123"
)
# Log a step
trace.log_step(
name="retrieve_documents",
duration_ms=245,
tokens_used=1024,
status="success"
)
# End trace
trace.end()
CLI Usage
# Start the monitoring server
aos-server --port 8000 --db-url postgresql://localhost/observability
# Stream live traces
aos-trace watch
# Export session data
aos-export --session session-123 --format json --output trace.json
# Analyze costs
aos-costs --agent-id my-agent --date-range "2025-03-01:2025-03-18"
Usage Examples
Monitor agent costs in real-time:
trace = client.start_trace(agent_id="researcher")
# ... agent work ...
cost_report = trace.get_cost_summary()
print(f"Total tokens: {cost_report.total_tokens}")
print(f"Estimated cost: ${cost_report.estimated_cost:.4f}")
Capture and replay failures:
failed_traces = client.query_traces(status="error", limit=10)
for trace in failed_traces:
print(f"Error: {trace.error_message}")
print(f"Decision tree: {trace.decision_path}")
Subscribe to live events:
async def handle_trace(event):
print(f"Trace {event.trace_id}: {event.status}")
client.subscribe("trace.completed", handle_trace)
Tech Stack
- Runtime: Python 3.12+
- API: FastAPI with async support
- Database: PostgreSQL with SQLAlchemy ORM
- Real-time: WebSocket streams via websockets library
- Configuration: Pydantic for settings management
- CLI: Click for command-line interface
- Packaging: Poetry for dependency management
Architecture
The studio consists of four main components:
- API Server (
api.py) — RESTful endpoints for trace queries and configuration - WebSocket Manager (
websocket_manager.py) — Real-time event streaming - Database Layer (
database.py) — Persistent storage of traces and metrics - Client SDK (
client.py) — Python library for agent instrumentation
License
MIT
Metadata
Release files for agent-observability-studio 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| agent_observability_studio-0.1.0.tar.gz | 9.3 kB | Without provenance | Details for agent_observability_studio-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agent_observability_studio-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for agent_observability_studio-0.1.0-py3-none-any.whl |
Total release size: 21.6 kB
Release files / agent_observability_studio-0.1.0.tar.gz
| Download URL | agent_observability_studio-0.1.0.tar.gz |
|---|---|
| Size | 9.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
224343038585852362854d3e0b9a014f9246c53e86c36f057ef9df306fb370d5
|
|
BLAKE2b-256 checksum How to use checksums |
afd87183eef8e933c88a5e633b1b39f6d06aaf03033cf6141268f13564b80831
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.9.25
|
Release files / agent_observability_studio-0.1.0-py3-none-any.whl
| Download URL | agent_observability_studio-0.1.0-py3-none-any.whl |
|---|---|
| Size | 12.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
6708a57ae0ba0b8889f57621351b239cb722ada6818501817b0712c0137bd921
|
|
BLAKE2b-256 checksum How to use checksums |
804f57f026976b6db0e532612d03f277c9f0e0a3ba6f1180ec5ec4dfdaa839b2
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.9.25
|
Security for agent-observability-studio 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.