p99.chat – Performance Optimization Playground

TL;DR

TL;DR: p99.chat is a free browser-based playground for measuring software latency, analyzing tail performance (p99/p95/p90), and running optimization experiments without installing anything.

Source and Accuracy Notes

• Official site: https://p99.chat/chat
• HN Show HN: https://news.ycombinator.com/item?id=65234510
• Creator: art049 (HN)

What Is p99.chat?

p99.chat is a web-based performance engineering tool that lets you measure, visualize, and optimize software latency characteristics directly in the browser. It focuses on tail latency analysis — the kind that matters for user-facing services where slow outliers tank the experience.

The core use case: paste code or run microbenchmarks, then inspect latency distributions, identify bottlenecks, and measure the impact of optimizations. No local setup, no install, no accounts required.

Setup Workflow

Step 1: Open the Playground

Navigate to https://p99.chat/chat. You’re dropped straight into the interface — no sign-up, no API key.

Step 2: Write or Paste Code

The editor supports multiple languages. Write a function you want to profile, or paste an existing snippet. The tool runs it in a sandboxed environment and collects timing data.

Step 3: Run the Experiment

Hit “Run” and p99.chat executes your code multiple times (typically hundreds to thousands of iterations) to build a latency distribution. It then surfaces:

• Mean, median, p50, p95, p99, p99.9 latency
• Histogram of response times
• Outlier analysis — which runs were slowest and why

Step 4: Iterate and Optimize

Modify your code and re-run to measure the delta. The interface shows before/after comparisons so you can quantify the impact of each change.

Deeper Analysis

What Makes p99 Special for Performance Work?

Traditional profiling tools give you CPU time and call counts. p99.chat gives you latency distributions — which is what actually matters for user-facing latency. A service that averages 50ms but has p99 of 800ms will tank under load even if ab shows “50ms average.”

The tool is particularly useful for:

• Comparing algorithms — O(n log n) vs O(n²) becomes viscerally obvious when you see the tail difference
• Testing optimization tricks — Does switching to memcpy actually help? Run it and see
• Understanding caching effects — Cold cache vs warm cache latency gaps are immediately visible
• Library evaluation — Benchmark a library’s actual performance profile before committing

Technical Limitations

• Sandboxed execution — Code runs in a constrained environment. No filesystem access, no network I/O beyond what the tool provides.
• Single-function scope — You can’t profile multi-service distributed systems. It’s scoped to microbenchmarks.
• No native code profiling — Works best for interpreted languages or code with FFI overhead that’s measurable in isolation.

Practical Evaluation Checklist

• [ ] Does it run code in the browser without an account?
• [ ] Are p50/p95/p99 metrics clearly displayed?
• [ ] Does it show latency histograms?
• [ ] Can you compare before/after runs?
• [ ] Is the interface responsive and usable on mobile?
• [ ] Are there pre-built example benchmarks to learn from?

Security Notes

• Code execution is sandboxed — no access to your filesystem or network
• No data persistence between sessions (stateless)
• No cookies or tracking beyond session cookies for UI state
• Works entirely client-side for most operations

FAQ

Q: Is p99.chat free? A: Yes, the core playground is free with no account required. There may be rate limits on very heavy workloads.

Q: What languages does it support? A: Primarily JavaScript/TypeScript for browser execution. The creator has indicated plans to expand language support.

Q: How is this different from jsperf or Benchmark.js? A: jsperf is focused on comparing two implementations. p99.chat is focused on measuring a single implementation’s latency distribution and understanding tail behavior — not just “which is faster” but “how much worse are the slow outliers.”

Q: Can I use this to profile production services? A: No — it’s a microbenchmarking tool scoped to single functions. For distributed tracing and production profiling, look at OpenTelemetry, Jaeger, or commercial tools like Datadog APM.

Q: Who built this? A: art049, an individual developer. The tool launched on Show HN in June 2025.

Conclusion

p99.chat fills a specific niche: quick, frictionless latency microbenchmarking in the browser. If you need to understand tail latency behavior of a function, algorithm, or library call — without installing any tooling — it’s a solid choice. The lack of sign-up and instant availability makes it ideal for one-off experiments.

Bookmark it next time you’re debating whether a “minor” optimization is worth the complexity. Run it, measure it, decide.

Next steps:

• https://p99.chat/chat — open the playground
• Read the HN discussion for context
• Try benchmarking a function you’re currently debating optimizing