Quick Start

Send your first agent trace to AgentVista in under 5 minutes.

Get your first trace into the dashboard by following these steps.

1. Sign up

   Create your account at agentvista.dev. The free tier gives you 3 agents and 500 runs per month — no credit card required.

2. Create an API key

   In the dashboard, go to Settings → API Keys → Create Key. Give it a name (e.g., dev) and select the write scope.

   Copy the key immediately — it is shown only once and cannot be retrieved later.

   Your key will look like av_a1b2c3d4... (64 hex characters after the prefix).

3. Install the SDK

   Base

   pip install agentvista
   
4. Initialize at startup

   Call agentvista.init() once when your application starts — before any agents run.

   import agentvista
   
   agentvista.init(api_key="av_a1b2c3d4...")
   # Or from an environment variable:
   # agentvista.init(api_key=os.environ["AGENTVISTA_API_KEY"])
   
5. Instrument your agent

   Add the @agentvista.trace_agent decorator to any function that runs your agent logic. Every call becomes a trace in the dashboard.

   import agentvista
   
   agentvista.init(api_key="av_a1b2c3d4...")
   
   @agentvista.trace_agent
   def classify_support_ticket(ticket: str) -> str:
       # Your existing agent code — no other changes needed
       response = call_your_llm(ticket)
       return response
   
   # Call it normally
   classify_support_ticket("Customer can't log in")
   

   If AgentVista is unreachable, your agent continues normally. The SDK drops events silently — it never blocks or crashes your application.

6. View your trace

   Open the AgentVista dashboard and navigate to Agents. Your agent appears automatically — click it to see the trace waterfall, timing, and status.

Set an outcome

Tell AgentVista whether a run succeeded. Use the run() context manager for explicit control:

import agentvista

agentvista.init(api_key="av_a1b2c3d4...")

def classify_support_ticket(ticket: str) -> dict:
    with agentvista.run("support-classifier") as r:
        result = call_your_llm(ticket)
        r.set_outcome(success=result["resolved"], outcome=result["category"])
        return result

Outcomes appear as a Success Rate chart per agent in the dashboard.

Next steps

Tracing

Explore decorators, context managers, child spans, and distributed tracing.

Anthropic Adapter

Auto-capture token counts and cost from Claude responses.

OpenAI Adapter

Auto-capture token counts and cost from OpenAI responses.

OpenTelemetry

Send infrastructure metrics and logs alongside your agent traces.