Traces & replay
Every LLM call, tool call, retrieval and sub-agent on one timeline, with the inputs, outputs, tokens and cost of each step. Replay any run with a new prompt or model.
What you get
Sub-agents and parallel tool calls nest under the step that spawned them, so the shape of the run is obvious.
Pretty JSON or rendered markdown for every span, with PII masked by your redaction rules.
Re-run from any span with a different prompt version, model or tool result, and diff the outcome.
Save a run as a test case in one click. Score every prompt change against real traffic.
Watch runs stream in while you deploy. Filter by agent, status, customer or trigger.
Bring spans from LangGraph, CrewAI or your own framework through any OTel exporter.
How it works
from fernline import trace
@trace(agent="refund-assistant")
def handle(ticket):
order = tools.lookup_order(ticket.customer)
return llm.decide_refund(order, policy)pip or npm, then wrap your agent entry point. Two lines.
Spans stream in as the run happens. Tools and MCP servers are detected automatically.
{
"customer": "#5086"
}Click any run to see its waterfall, then any span to see what went in and what came out.
Customers
“The trace view is the first thing I open in an incident. Time to root cause went from an afternoon to a coffee.”
FAQ
Can't find what you're looking for? Write to us and a human answers within a day.
Free for your first 100k spans. Set up in ten minutes, see your first trace in two.
No card needed · SOC 2 Type II · Cancel anytime