ClawSight - AI Agent Observability Dashboard

TL;DR

TL;DR: ClawSight is an open-source observability platform for AI agents — real-time logging, token usage tracking, and kill switches for runaway bots, deployable in 2 minutes.

Source and Accuracy Notes

• Website: clawsight.org
• GitHub: cantinaxyz/clawsight-siem
• License: Open source
• Stack: Node.js

What Is ClawSight?

ClawSight is an open-source observability platform purpose-built for autonomous AI agents. It gives you a unified dashboard to monitor costs, read execution logs, and terminate runaway bot processes before they burn through your budget.

The core problem it solves: once you deploy multiple AI agents in parallel, you have no visibility into what each one is doing, how many tokens it’s consuming, or whether any agent has gone off the rails. ClawSight makes that observable.

Setup Workflow

Step 1: Prerequisites

• Node.js 18 or later
• npm or yarn

Step 2: Install

npm install -g clawsight

Step 3: Configure Your Agent

Point your agent to the ClawSight endpoint by setting an environment variable:

export CLAWSIGHT_ENDPOINT="http://localhost:3000"
export CLAWSIGHT_API_KEY="your-api-key"

Step 4: Start the Dashboard

npx clawsight start

Open http://localhost:3000 to access the dashboard. You’ll see active agents, token usage charts, and a live log stream.

Step 5: Integrate an Agent

For a simple Node.js agent integration:

import { ClawSight } from 'clawsight';

const monitor = new ClawSight({
  endpoint: process.env.CLAWSIGHT_ENDPOINT,
  apiKey: process.env.CLAWSIGHT_API_KEY,
});

monitor.start();

Deeper Analysis

What makes it different: Most agent observability tools are enterprise SaaS with opaque pricing. ClawSight is self-hostable and gives you the core triad — logs, costs, kill switch — without a sales call.

Token analytics: The dashboard breaks down token consumption per agent, per conversation, and per time window. Useful for catching agents that are prompting excessively.

Kill switch: The remote termination feature lets you stop a running agent process by ID. No need to dig into container logs or SSH into a server.

Limitations: Currently focused on Node.js agents. Python and other runtimes are not yet supported out of the box.

Practical Evaluation Checklist

• Open clawsight.org and check the live demo dashboard
• Review the GitHub repo for setup documentation
• Run npx clawsight start locally to test the dashboard
• Check if your agent runtime is supported (Node.js required)
• Evaluate whether the token analytics meet your needs

Security Notes

• Run ClawSight behind your firewall for internal agents handling sensitive data
• API keys should be treated as secrets — rotate them regularly
• The kill switch feature has no per-user authorization by default — restrict network access to the dashboard

FAQ

Q: Is ClawSight free? A: Yes, it is fully open source and self-hostable at no cost. There is no enterprise tier or licensing fee.

Q: What runtimes does it support? A: Currently Node.js agents. Python support is on the roadmap.

Q: Can I self-host it? A: Yes. ClawSight runs as a Node.js service you deploy on your own infrastructure — cloud, on-prem, or local machine.

Q: How does the kill switch work? A: The dashboard communicates with a lightweight agent-side library that receives termination signals. When you click “stop” on an agent in the UI, the running process is killed remotely.

Conclusion

ClawSight fills a real gap in the AI agent tooling stack: observability without vendor lock-in. If you’re running multiple agents and flying blind, it takes 2 minutes to get visibility. The open-source model means you can audit what it collects and self-host wherever you run your agents.