47
runs last night. Three failed. You found out this morning.
v1 · Agent observability
AI agents fail silently and bill loudly.
Most developers find out their agent broke from an angry user. Farol tells you first.
Open source SDK · View on GitHub →
The gap
You can’t fix what you never see
3×
higher token costs after a prompt tweak last Tuesday. You noticed on Friday.
1
decorator to get started
sdk.py
from farol import trace
@trace(agent_name="my-agent", farol_key="frl_...")
def my_agent(task, run=None):
run["topic"] = task # that’s itProduct
Farol watches your agents so you don't have to.
Never get surprised by a bill again
Farol tracks token spend per run and learns what's normal for each agent. The moment something costs 3× more than usual, you get an alert — before your cloud provider does. Set monthly budget limits and get notified the instant they're crossed.
Your agent broke. Find out why in seconds.
Open any run and see every tool call, LLM call, duration, and error reconstructed end to end. No log diving, no guessing. The trace timeline shows exactly where things went wrong and how long each step took.
Know if your agents are healthy without opening the dashboard.
A weekly email lands every Monday with your agents' success rate, cost, and quality score for the week. Green means sleep. Red means act. One email, no noise.
The only monitoring tool that tracks whether your agent was actually good.
Rate outputs with a thumbs up or down. Farol tracks quality over time and alerts you when it drops — so you catch prompt degradation before your users do.
Your agent started failing more often. Farol noticed before your users did.
Automatic regression detection compares this week's success rate against last week's. If something changed, you'll know — with which agent, how much, and a direct link to investigate.
See your entire agent pipeline in one place.
Link child agents to their parent run and see total pipeline cost, duration, and errors across the entire chain. Know exactly which agent in a multi-step pipeline caused the problem.
Get warned before things break — not after.
Farol detects when your agent is trending slower, more expensive, or less reliable — even when nothing has technically failed yet. Catch drift early, before it becomes an incident.
Show your agent's performance to anyone, no login required.
Generate a read-only dashboard link and share it with clients, investors, or teammates. They see live data. They don't see your settings or API keys.
Know if your agent's outputs are actually correct.
Run LLM-as-judge evaluations on any trace. Score for correctness, groundedness, JSON validity, and tone. See pass rates trend over time — not just thumbs up or down.
Connect to your existing observability stack.
Point your OpenTelemetry exporter at Farol and start seeing traces instantly. No SDK changes required. Works with any OTLP/HTTP compatible tool.
ALWAYS ON
While you focus on building, Farol is watching.
- Alerting you when a run costs 3× more than usual — before your cloud bill arrives
- Detecting when your agent's success rate drops week over week
- Watching output quality and alerting you when it degrades
- Tracking your Agent Health Score (0-100) and alerting you when it drops week over week
- Grouping related alerts when multiple agents are affected by the same issue
- Sending you a Monday morning summary of every agent's health, cost, and quality
- Scoring your agent outputs automatically with LLM-as-judge evals
- Accepting OpenTelemetry traces from any OTLP-compatible exporter
Pricing
Pick a plan that scales with your agents
Simple pricing. No sales call.
Free
€0/mo
- 1 agent
- 50k events/month
- Cost anomaly detection
- Full trace inspector
- Email alerts
- Agent Health Score
Starter
€20/month
excl. VAT
- 3 agents
- 300k events/month
- Everything in Free
- CSV export
- Shared read-only dashboard
- p95 latency tracking
- Run notes & quality scoring
- Agent Health Score
- Prompt version tagging
Most popular
Builder
€50/month
excl. VAT
- Unlimited agents
- Unlimited events
- Everything in Starter
- Native Slack + webhook alerts
- Budget alerts per agent
- Weekly digest email
- Regression & quality trend alerts
- Proactive trend alerts
- Alert grouping
- Multi-agent trace linking
- Prompt version comparison
- Pipeline DAG visualization
- Evals system (LLM-as-judge)
Studio
€199/month
- Everything in Builder
- 5 team seats
- Early access to new features