v0.8 — open beta, SOC 2 in progress

Production monitoring for AI agents.

Catch agent drift before your users do. Detect silent regressions across every session, tool call, and LLM interaction — without writing a single threshold.

Start free Book a Demo

$ pip install dottle·2 lines to instrument·Works with any framework

Trusted by teams shipping agents atNORTHPIKELumen/AIHARBORaxiom.soMeridianstackline

Why dottle

We don’t just show logs. We tell you what broke.

Other tools give you observability — a passive stream of data you still have to interpret. Dottle does the analysis for you.

Detect

Detect failures automatically

Loops, tool errors, bad outputs — caught the moment they happen. No manual thresholds or configs.

Trace

Trace every agent run

Every step, tool call, and LLM decision captured end-to-end. Full timeline, not just error lines.

Isolate

Highlight exactly where it failed

Not "something broke at step 4." The exact call, input, and response — isolated and explained.

Diagnose

Suggest the root cause

Pattern-matched diagnosis across thousands of agent runs. We tell you why and how to fix it.

Experiment

A/B test prompts in production

Compare prompts, models, and configs side-by-side with statistical rigor. Know what actually works.

Alert

Slack the right person, fast

Real-time alerts when agents spike errors or drift. Know about issues before users report them.

How it works

Two lines. Every step captured.

Drop the SDK in. Dottle instruments your agent’s tools, LLM calls, and control flow automatically. No custom spans, no manual thresholds.

01Install the SDK
02Wrap your agent entrypoint
03Ship to prod — dottle watches from step 1

1from dottle import Dottle
2from openai import OpenAI
3
4dottle = Dottle(project="support-agent")
5
6@dottle.trace
7def run_agent(user_msg: str):
8    # Dottle auto-captures every tool call,
9    # LLM response, and control flow decision
10    return agent.invoke(user_msg)
11
12# That's it. Drift, loops, and silent
13# failures are flagged automatically.