// docs
Get started in 60 seconds
Install the SDK, drop your API key, send a trace. View it in the dashboard.
1. Install the SDK
Python 3.9 or later.
bash
pip install agentlogs-sdk2. Get your API key
- 1. Sign in with email (magic link).
- 2. Go to Settings → Projects.
- 3. Create a project + an API key. Copy the key (you'll only see it once).
Then set it in your environment:
bash
export AGENTLOGS_API_KEY="al_..."
export AGENTLOGS_PROJECT="my-agent"3. Send your first trace
python
from agentlogs import AgentLogs, StepType
client = AgentLogs(api_key="al_...") # or set AGENTLOGS_API_KEY env var
async with await client.create_trace(name="research-agent") as trace:
await trace.add_step(
name="openai_call",
step_type=StepType.LLM,
metadata={"model": "gpt-4o", "provider": "openai"},
tokens=1234,
cost=0.012,
)Open the dashboard — your trace appears within seconds.
Decorator API (alternative)
Auto-trace an async function:
python
from agentlogs import trace_agent
@trace_agent()
async def my_agent(query: str):
return await llm.generate(query)Step types
python
from agentlogs import StepType
StepType.LLM # OpenAI / Anthropic / etc. calls
StepType.TOOL # External tool / function call
StepType.FUNCTION # Internal function step
StepType.OTHER # Anything elseREST API reference
All SDK calls go through these HTTP endpoints. Use directly from any language.
POST
/api/v1/tracesCreate a new trace. Returns trace id.
{
"name": "agent-name",
"input_data": { ... }, // optional
"metadata": { ... }, // optional
"tags": ["a", "b"] // optional
}POST
/api/v1/traces/:id/stepsAppend a step to a trace.
{
"name": "step-name",
"type": "llm|tool|function|other",
"status": "success|error",
"input": { ... },
"output": { ... },
"metadata": { ... },
"tokens": 1234,
"cost": 0.012,
"duration_ms": 450
}PATCH
/api/v1/traces/:idFinalize a trace.
{
"status": "success|error",
"output_data": { ... },
"error_message": "...",
"total_tokens": 5000,
"total_cost": 0.045
}Authorize each request with your API key:
bash
curl -X POST https://agentlogs.app/api/v1/traces \
-H "Authorization: Bearer $AGENTLOGS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"name": "my-agent-run", "metadata": {"version": "1.0"}}'Self-hosting
Point the SDK at your own deployment by overriding the API URL:
python
from agentlogs import configure
configure(api_url="https://logs.your-company.com")Source code: github.com/infosiva/agenttrace