Open Source·Self-Hostable

Agent observability built for agents
— not retrofitted from LLM logging.

Real-time trace view, issue lifecycle tracking, and step-by-step execution— for AI agents that actually run.

agentlogs.app — live trace streamLIVE
14:32:01.423[INFO]research-agentStarting task: "Summarize latest AI papers"
14:32:01.891[CALL]research-agenttool_call: search_arxiv(query="transformer architecture 2024")
14:32:02.340[INFO]research-agentRetrieved 47 papers, filtering by relevance score > 0.85
14:32:02.891[CALL]research-agentllm_call: claude-3-5-sonnet (tokens: 4,291)
14:32:04.102[INFO]research-agentStep completed: latency=1.2s cost=$0.0043
14:32:04.201[WARN]summarizerRate limit approaching: 85% of quota used
14:32:04.890[CALL]summarizertool_call: fetch_url(url="https://arxiv.org/abs/2401.12345")
14:32:05.112[INFO]summarizerChunk 1/3 processed: 2,048 tokens compressed to 312
14:32:05.445[INFO]summarizerChunk 2/3 processed: 1,891 tokens compressed to 287
14:32:05.901[INFO]summarizerChunk 3/3 processed: 2,143 tokens compressed to 321
14:32:06.003[DONE]research-agentTask complete — total: 3.58s / $0.0124 / 5 steps
14:32:06.244[INFO]monitorTrace saved: trace_7f3a9b2c [id: 7f3a9b2c]
14:32:07.001[INFO]writer-agentStarting task: "Draft blog post from summary"
14:32:07.445[CALL]writer-agentllm_call: gpt-4o (tokens: 1,102)
14:32:08.330[ERROR]writer-agentOpenAI timeout after 1000ms — retrying with fallback
14:32:08.891[CALL]writer-agentllm_call: claude-3-haiku (fallback, tokens: 1,102)
14:32:09.210[DONE]writer-agentDraft complete — 847 words / latency 1.78s / $0.0031
sample dashboard output
Req/sec
2,847
+12%24h
P99 Latency
1.24s
-8ms24h
Error Rate
0.3%
-0.1%24h
Cost/1k
$0.041
-4%24h
// capabilities

Everything you need to ship reliable AI

Built-in observability. Zero vendor lock-in. Works with any framework.

Real-time Tracing
Watch agents think step-by-step with full execution visibility
Cost Analytics
Track token usage and API costs across every agent run
Latency Profiling
Find bottlenecks. P50/P95/P99 breakdowns per step
Error Detection
Auto-categorize failures. Alert before users notice
Multi-Framework
LangChain, CrewAI, AutoGPT, or your own custom agents
Self-Hostable
Your data never leaves your infra. Docker in 2 minutes
SDK First
Python + TypeScript. 3 lines to full instrumentation
Production Scale
Built on FastAPI + Postgres. Async event ingestion designed to scale
// quickstart

Instrument in 60 seconds

01Install SDK
pip install agentlogs-sdk
02Set endpoint
export AGENTLOGS_API_KEY="<your API key>"
export AGENTLOGS_PROJECT="my-agent"
03Instrument
from agentlogs import AgentLogs

client = AgentLogs()
with client.trace("agent-run") as trace:
    trace.step("llm_call", metadata={"model": "gpt-4o", "tokens": 1200})
    # your agent logic
    trace.step("done", metadata={"output": result})
Get your API key →Free, no card required
// pricing

Simple. Developer-first.

Start free. Upgrade when you need more.

Free
$0
Forever free, open source
  • 10,000 trace events / month
  • 7-day log retention
  • 1 project
  • Community support
Get started free
POPULAR
Pro
$19
per month, billed monthly
  • 1,000,000 trace events / month
  • 30-day log retention
  • Unlimited projects
  • Team seats (up to 5)
  • Webhook + REST API export
  • Email support (24h SLA)
// compare

How we stack up

FeatureAgentLogsLangSmithHeliconeArize Phoenix
Full Step-Tree Tracing
Replay / Root-Cause Debug
OTel gen_ai Ingest
Cost Analytics
Self-Hosting
Open Source
Starting PriceFree$39/seat/mo$79/moFree (OSS)

Pricing as of Sept 2026, publicly listed rates.

$ ready to deploy

Ship AI agents with confidence

Open-source observability for AI agents. Free forever for the core. Self-host or use our hosted version.

Start Free Trial →