Open source · Self-hostable · Anthropic native

See exactly what your
AI agent is doing

Argus gives engineering teams real-time visibility into every Claude API call — what it cost, how long it took, what failed, and why. Know what your AI is doing in production before your users notice something is wrong.

< 2ms

Tracing overhead

4

Models supported

100%

Open source

Three steps to full visibility

No infrastructure changes. Works with your existing Claude setup.

01

Wrap your client

Replace anthropic.Anthropic() with ArgusClient(). One line change, nothing else breaks.

from argus import ArgusClient

client = ArgusClient(session_label="My App")
02

Use it normally

Call the API exactly as you always have. Argus captures everything silently in the background.

response = client.messages.create(
  model="claude-opus-4-8",
  messages=[...]
)
03

Open the dashboard

See every call, cost, latency, and error in real time. No configuration needed.

# Open your dashboard
# Every trace appears instantly

Everything you need to understand your agent

Purpose-built for teams running Claude in production — no wrappers, no vendor lock-in.

Real-time traces

Every API call is captured the moment it happens. Watch your agent work live, call by call.

Cost tracking

Know exactly what every session costs. Per-call breakdown and cumulative session totals across all models.

Latency monitoring

Track response times per call and per model. Identify which parts of your agent are slow.

Error visibility

Every failure is logged with type, message, and context. Stop guessing why your agent broke in production.

Start tracing in under a minute

Self-host with Docker Compose or connect to any running instance.