Coding harness usage signals
Inspect source-separated activity by signal, harness, or focused comparison. Every view preserves its own unit, window, coverage, and limitations.
- Different units
- Signals remain separate and cannot be added.
- Missing is not zero
- Unmapped coverage stays explicitly unknown.
- Context, not quality
- Activity never changes capability or classification.
Most used on OpenRouter
Traffic attributed to public coding apps using OpenRouter's app-attribution mechanism.
OpenRouter controls
Linear bars use the largest mapped value in this source as 100%. Positive values below one display pixel retain a 1 px origin marker. Exact values remain authoritative.
- #1Hermes AgentPersistent open agent with coding tools, durable automation, and isolated runtimes.79,269,896Requests | Hermes Agent
- #2Claude CodePolished Claude-first agent across terminal, IDE, desktop, and web.18,873,021Requests | Claude Code
- #3Kilo CodeOne open agent across IDE, terminal, cloud, and CI.23,059,474Requests | Kilo Code
- #4OpenClawAlways-on personal agent harness with coding tools, durable automation, and broad model choice.20,914,077Requests | OpenClaw
- #5Pi AgentMinimal terminal harness designed to be extended, embedded, or automated.14,363,889Requests | pi
- #6ClineHuman-in-the-loop agent across IDE, terminal, and teams.6,005,044Requests | Cline
- #7CodexAgentic coding across app, terminal, IDE, cloud, and automation.2,527,712Requests | Codex
- #9CodebuffAutonomous multi-agent coding from the terminal or TypeScript.5,271,085Requests | Codebuff
- #10Oh My PiA batteries-included Pi fork with LSP, debugger, browser, and subagents.1,536,902Requests | Oh-My-Pi
- #13Cursor CLICursor's familiar agent loop in the terminal and in automation.818,357Requests | Cursor
- #16OpenHandsAgent platform with selectable local, sandboxed, and remote runtimes.1,015,478Requests | OpenHands
- #19MiMo CodeOpen coding harness with persistent memory and structured multi-agent workflows.218,234Requests | mimocode
How to read the source viewsUnits, coverage, and source populations
Why these sources stay separate
OpenRouter tokens, Homebrew install events, package downloads, editor installs, and repository activity observe different populations. Adding them together would create a precise-looking but invalid total.
- Routing
- OpenRouter attributed tokens and requests over completed 1-day, 7-day, and 30-day UTC windows.
- Distribution
- Homebrew 30d events, npm last-month downloads, and cumulative downloads of explicitly matched stable GitHub release assets.
- Editor marketplaces
- Cumulative VS Code installs, Open VSX downloads, and JetBrains downloads for exact mapped extensions or plugins.
- Repository interest
- GitHub stars for the repository already audited by HarnessMatch, with full, client, or support scope preserved.
- Visual scale
- Every bar uses a linear zero-based scale where the largest mapped value in the selected source is 100%. Positive subpixel values retain a 1 px origin marker; exact labels remain authoritative.