🚀 TL;DR

ClawTrace is the observability platform for OpenClaw agents. It captures every trajectory your OpenClaw agent runs — every LLM call, tool use, and sub-agent delegation — and gives you three views to understand it: an execution path, a call graph, and a timeline. When you need answers fast, ask Tracy, your built-in AI analyst, in plain English. Watch our demo here:

—

Hello, world! I'm Richard, and today I'm excited to introduce ClawTrace — the observability platform for OpenClaw agents.

AI agents are going to production. But right now, when something goes wrong, you're flying blind.

❌ The Problem

Deploying an OpenClaw agent is easy. Understanding what it actually did is not.

When a run costs $104.70 instead of the usual $1.5, you have no idea why. When an agent fails silently, you don't know which step broke or what input caused it. When a trajectory takes 3 minutes instead of 30 seconds, you can't tell where the time went. You end up reading raw JSON logs, manually counting token estimates, and guessing at execution paths. This doesn't scale.

Existing options make this worse, not better:

• General distributed tracing tools (Jaeger, Zipkin, Datadog) weren't built for LLM concepts — they have no understanding of token counts, model costs, prompt/completion pairs, or sub-agent hierarchies.
• Framework-specific tools like LangSmith only work within their own ecosystem. If you're on OpenClaw, you're on your own.
• Point solutions cover one slice of the problem — just logs, just cost tracking, just evals — and leave you to stitch everything together yourself.
• Every tool dumps data into a dashboard and expects you to interpret it. There's no analyst. You still have to do all the thinking.

⭐️ Our Solution

We built ClawTrace to give you complete visibility into every OpenClaw agent run — automatically, with zero code changes to your agent.

One plugin. Three commands. Every trajectory captured.

openclaw plugins install @epsilla/clawtrace
openclaw clawtrace setup
openclaw gateway restart

That's it. From that point on, every agent run streams to ClawTrace automatically.

With ClawTrace, you get:

• Execution Path — a collapsible tree of every step your agent took: LLM calls, tool executions, sub-agent delegations, with full input and output payloads at every node. Metadata badges show model name, duration, token counts, and estimated cost inline.

  

• Call Graph — an interactive force-directed diagram showing the relationships between every agent, model, and tool in a run. Understand the shape of a complex multi-agent workflow at a glance.

  

• Timeline — a Gantt chart of every span in the trajectory, positioned by start time and sized by duration. Bottlenecks, parallelism, and idle gaps become immediately visible.

  

• Tracy, your OpenClaw’s Doctor — every other observability tool gives you data and expects you to figure it out yourself. Tracy is different. Ask her a question in plain English — "where is the bottleneck in this trace?" — and she queries your trajectory graph live, generates a visualization, and delivers specific recommendations: which step to fix, why, and how. Tracy is powered by a Claude managed agent and has real-time Cypher access to your full trajectory graph via PuppyGraph graph lake-house architecture.

  

• Cost Estimation — per-span cost calculation covering 80+ models across OpenAI, Anthropic, Google, DeepSeek, Mistral, Alibaba Qwen, Zhipu GLM, Moonshot, and more. Cache-aware pricing (fresh input vs. cached input vs. output tokens) for accurate estimates.

• Consumption-based billing — pay for what you use with credits. No seat fees, no per-user pricing. Start with 100 free credits.

🔦 The Ask

AI agents are moving from demos to production. ClawTrace is the observability layer that makes that transition safe, debuggable, and cost-efficient.

• 🚀 Sign up for free at clawtrace.ai — 100 credits included, your first trajectory shows up in minutes.
• ⭐ Star us on GitHub — ClawTrace is fully open source under Apache 2.0: github.com/epsilla-cloud/clawtrace
• 💬 Tell us what you're building — we'd love to hear about your OpenClaw setup and what observability problems matter most to you. Schedule a call with me.