Demo // How It Works

Monitor your agents
in 60 seconds

Install the SDK, instrument your code, and watch live traces appear in your dashboard — no backend setup required.

pip install agentwatch OpenAI · CrewAI · LangGraph Self-hosted
// LIVE TRACE streaming
Install SDK 0.3s
Add decorator 1.2s
Run agent 4.8s
Trace arrives —
847 tokens
$0.12 cost
3 tool calls
// OPERATION: AGENTWATCH — STEP BY STEP
STEP 01
Install the SDK
Pull in the AgentWatch package and any framework extras you need.
# Core package pip install agentwatch # Framework extras pip install agentwatch[openai] pip install agentwatch[crewai] pip install agentwatch[langgraph]
STEP 02
Configure your environment
Point the SDK at your AgentWatch deployment with one environment variable.
# .env AGENTWATCH_API_URL=https://agentwatch-3.polsia.app # Optional: cap per-session spend in cents AGENTWATCH_MAX_COST_CENTS=500
STEP 03 — A
Instrument with a decorator
Wrap any async function with @aw.track() — the session opens, runs, and uploads automatically.
from agentwatch import AgentWatch aw = AgentWatch() # Decorator — session auto-managed @aw.track() async def my_agent(task: str): result = await llm.ainvoke(task) return result
STEP 03 — B
Instrument with a context manager
Use with aw.session() for a named session with explicit scope boundaries.
from agentwatch import AgentWatch aw = AgentWatch() # Context manager — named session with aw.session(name="order-processing") as session: # ... agent runs here ... session.record_llm_call( model="gpt-4o", input_tokens=150, output_tokens=340 ) # session auto-closes and uploads trace
STEP 03 — C
Instrument with the manual API
Start a session, record events individually, then close — full control for advanced use cases.
from agentwatch import AgentWatch aw = AgentWatch() session_id = aw.start_session(name="research") aw.record_llm_call( model="gpt-4o", input_tokens=150, output_tokens=340 ) aw.record_tool_call( tool="web_search", latency_ms=230 ) aw.end_session()
STEP 04
Observe in the dashboard
Open the AgentWatch dashboard to explore traces, costs, and governance alerts in real time.
# Your deployed dashboard https://agentwatch-3.polsia.app/sessions # Per-session trace detail https://agentwatch-3.polsia.app/sessions/:id # Governance alerts https://agentwatch-3.polsia.app/alerts
// SUPPORTED FRAMEWORKS
OpenAI
Agents SDK
First-class support via the agents SDK instrumentation layer built into agentwatch.
CrewAI
Capture every crew execution, task delegation, and tool invocation across your agents.
LangGraph
Trace state graph updates, node execution times, and conditional branching in real time.
LlamaIndex
Capture context retrieval events, query_engine calls, and LLM FUNCTION_CALLS automatically.
AutoGen
Patch async reply generation hooks to capture agent-to-agent messaging and response deltas.
Start monitoring your agents today.
pip install agentwatch View Dashboard