Agent-native observability
Track tool calls, handoffs, memory reads, and decision trees — not just LLM token counts.
AgentLens
Unlike LangSmith (paid, cloud-only) and Langfuse (LLM-focused), AgentLens understands agents — tool calls, handoffs, memory reads, and decision trees.
Why AgentLens?
Section titled “Why AgentLens?”Self-hosted
docker run and done. Your traces never leave your infrastructure.
Live streaming
Watch your agent think in real-time with incremental span updates via SSE.
Multi-framework
LangChain, CrewAI, AutoGen, LlamaIndex, Google ADK, and any OTel-instrumented app.
Features at a Glance
Section titled “Features at a Glance”| Feature | Description |
|---|---|
| Live trace streaming | Watch your agent think in real-time with incremental span updates |
| Agent topology graph | Visualize agent spawns, tool calls, and handoffs as an interactive DAG |
| Trace comparison | Side-by-side diff of two agent runs with color-coded span matching |
| Search & filters | Full-text search, status/agent/date/cost filters, sortable columns, pagination |
| Cost tracking | 27 models priced — GPT-4.1, Claude 4, Gemini 2.0, DeepSeek, Llama 3.3 |
| Time-travel replay | Scrub through any trace step-by-step at any speed |
| Alerting | Detect cost spikes, error rates, latency anomalies, and missing spans |
| Multi-tenant auth | JWT sessions, API key auth, per-user data isolation |
| OpenTelemetry | Export to any OTel backend or receive OTLP HTTP JSON spans |
| 231 tests | Server + SDK with 100% coverage |
vs. Alternatives
Section titled “vs. Alternatives”| AgentLens | LangSmith | Langfuse | |
|---|---|---|---|
| Self-hosted | ✅ | ❌ cloud-only | ✅ |
| Agent topology graph | ✅ | ❌ | ❌ |
| Free | ✅ MIT | ❌ paid tiers | ⚠️ limited free |
| OTel ingestion | ✅ | ❌ | ✅ |
| Time-travel replay | ✅ | ❌ | ❌ |
| Alerting | ✅ | ❌ | ⚠️ enterprise |