- NewOpenTelemetry bridge (beta) — point your OTLP/HTTP exporter at Farol, no SDK changes required
- NewEvals system — run LLM-as-judge evaluations on agent outputs; score for correctness, groundedness, JSON validity, and tone
- NewEvals page (/evals) — configure eval rules per agent, run on any trace, view pass rate and score history
- NewEvals tab in trace modal — see per-trace eval results with score, pass/fail, and LLM reason inline
Changelog
Everything new, improved, and fixed in Farol.
- NewAutomatic trace propagation — parent agent runs now automatically link to child agents without manual parent_trace_id passing
- NewPipeline DAG visualization — multi-agent trace modal now shows a visual pipeline graph with per-agent cost, duration, and status
- NewCost optimization insights — detects redundant tool calls, model mismatch, and cost spikes with estimated monthly savings in the agent modal
- ImprovedPython SDK 0.1.7 — uses contextvars for async-safe context propagation
- ImprovedNode.js SDK 0.1.7 — uses AsyncLocalStorage for async-safe context propagation
- NewProactive trend alerts — get notified when cost, latency, or success rate is drifting before it becomes a problem
- ImprovedStatistical thresholds raised to 10+ runs minimum for more reliable anomaly detection
- ImprovedServer-side anomaly detection in ingest — consistent regardless of SDK version
- ImprovedPython SDK 0.1.6 — improved baseline detection
- ImprovedNode.js SDK 0.1.6 — version sync
- NewMulti-agent trace linking — pass parent_trace_id to connect child agent runs to their parent
- NewPipeline cost rollup — see total cost across all linked agents in the trace modal
- NewSpawned runs visible in trace modal — click through to any child or parent run
- ImprovedPython SDK 0.1.5 — add parent_trace_id parameter
- ImprovedNode.js SDK 0.1.5 — add parentTraceId parameter
- NewPrompt version tagging — tag runs with a version label (v2, after_refactor) to track how prompt changes affect cost and quality
- NewVersion comparison table in agent modal — see success rate, avg cost, and quality score side by side across prompt versions
- ImprovedPython SDK 0.1.4 — add prompt_version parameter
- ImprovedNode.js SDK 0.1.4 — add promptVersion parameter
- NewAgent Health Score — one number (0-100) combining success rate, cost, quality, and latency for each agent
- NewDate range filter on runs table — filter by Today, Last 7 days, Last 30 days, Last 90 days, All time
- NewRun search now includes agent name alongside topic and error
- Newp95 latency anomaly alerts — get alerted when tail latency worsens week over week
- NewDuration anomaly alerts — get alerted when a run takes 3× longer than your baseline
- NewAll/Unread filter on alerts — see only what you haven’t actioned yet
- ImprovedAlerts persist per user across logout/login
- NewNative Slack integration — send alerts directly to any Slack channel
- NewFeedback page — share bugs, feature requests, and general feedback at /feedback
- NewStudio plan — Coming Soon with notify me link
- ImprovedBuilder plan — €69/mo with unlimited events
- ImprovedPricing cards show specific features per tier
- NewPolar.sh billing — cleaner checkout, better developer experience
- NewQuality trend alerts — get alerted when output quality drops week over week
- NewRun notes — annotate any run with context directly from the trace modal
- NewError message search — search runs by topic, agent name, or error message
- NewBudget alerts — set monthly spend limits per agent, get notified when crossed
- NewWeekly digest email — every Monday, a summary of your agents’ health and cost
- NewShared read-only dashboards — share a link with clients or investors, no login required
- NewCSV export — download your runs filtered by agent and date
- NewQuality scoring — rate agent outputs thumbs up/down, track trends over time
- NewNode.js SDK — npm install @usefarol/sdk, full TypeScript support
- NewSampling support — send only a fraction of runs to reduce noise and costs
- NewWebhook alerts — send alerts to Discord, Google Chat, Zapier, or any HTTP endpoint
- ImprovedDocs — spans reference, updated SDK examples, FAQ expanded
- NewRegression alerts — automatic detection when success rate drops week over week
- Newp95 / p99 latency tracking — see your slowest runs, not just averages
- NewCost anomaly detection — baseline learning per agent, alerts on spikes
- NewFull trace inspector — every tool call, LLM call, and error reconstructed end to end
- NewPython SDK — pip install farol-sdk, one decorator to get started