See what your AI coding setup costs to run — and what is quietly drifting.
HarnessFit (powered by BHEB) audits any repository in under 2 seconds. Measure your hidden startup token tax, catch stale rules and broken Cloud VM links, and compare against peer setups — right from your browser.
projects/ folder). Nothing leaves your laptop.See exact eager startup tokens loaded on every turn and projected spend per 1,000 turns across DS4, MiniMax M3, Kimi, GLM, Flash, Haiku, and 470+ OpenRouter models.
Catch real production harness drift: stale .claude/SYNC_STAMP mirrors, OS-hardcoded paths breaking Linux Cloud VMs, missing test nets, and divergent AGENTS.md rules.
Default scans use 0 LLM tokens and upload nothing. Unlock deeper fixes & saved history via passwordless Email Magic Link only when you choose.
Scan a public GitHub URL, pick a local repository folder, or open our anonymized Live Sample Report to explore Cost to Operate, Dual Sub-Scores, and Harness Tax.
Compare Cost Across 8 Models (incl. MiniMax M3, GLM, Kimi, DS4) + Search 470+ OpenRouter ModelsAdd up to 5 Custom
6-Dimension Cohort Baseline ComparisonExpand Details
What Was Measured Here (5-Vector Technical Facts)Expand Facts
How LLMs Are Accessed & Exact Token Footprint Reassurance0 Tokens Default
AGENTS.md/CLAUDE.md; you never need Bot-Envy or Hermes installed.
claude/agy/codex CLI, or repo-check (0 tokens).OPENROUTER, ANTHROPIC, OPENAI, or GEMINI_API_KEY.--live-bench local-cli (your CLI), --live-bench byok (your API key), or --live-bench repo-check (0-token repo verification check).
Compare scores, Archetypes, startup token overhead, and model costs across peer setups without exposing private repo or builder identities.
cd into the repository you want to score and run any option below. No source code, prompts, or secrets ever leave your machine.
harnessfit_report.html locally.[Y/n] permission before checking global ~/.cursor, ~/.claude, and ~/.gemini skill folders.[Y/n] before creating your shareable web URL.Option 1 — Instant CLI Run via curl (macOS / Linux / WSL — zero install):
curl -fsSL https://harnessfit-behb.web.app/run.sh | bash
Option 2 — Launch Local Browser Testing UI (127.0.0.1 with Live Console, Explicit Permission Gate & Red Abort Button):
curl -fsSL https://harnessfit-behb.web.app/run.sh | bash -s -- ui
Option 3 — Install bheb / harnessfit CLI globally (then run harnessfit or harnessfit ui inside any repo):
curl -fsSL https://harnessfit-behb.web.app/install.sh | bash harnessfit
Include your Email immediately for a Magic Link (bookmark all past runs & re-runs):
curl -fsSL https://harnessfit-behb.web.app/run.sh | bash -s -- --email you@example.com
Offline-Only Mode (skip sharing prompt & keep 100% local):
curl -fsSL https://harnessfit-behb.web.app/run.sh | bash -s -- --no-host --share no
--repo .Scans the current repository (default when no subcommand is passed)--include-host/--no-hostControls checking~/.cursor,~/.claude,~/.gemini(prompts[Y/n]by default)--share ask|yes|noControls uploading anonymized stats for a web URL (prompts[Y/n]by default)--email you@example.comBinds uploaded scans to your personal Magic Link dashboard--live-bench no|local-cli|byok|repo-checkRun ~2.5k-token probe with your CLI/API key or 0-token repo check (Sponsored pool: Roadmap)--invite-repo <slug>Links your local diagnostic scan back to an invited repo report
AGENTS.md/CLAUDE.md; you never need Bot-Envy or Hermes installed.
claude/agy/codex CLI, or repo-check (0 tokens).OPENROUTER, ANTHROPIC, OPENAI, or GEMINI_API_KEY.--live-bench local-cli (your CLI), --live-bench byok (your API key), or --live-bench repo-check (0-token repo verification check).
Enter your email address to generate your personal bookmarkable Magic Link, view every repository and local host scan you have shared, and copy pre-filled re-scan commands.
Enter your email above (or open a ?magic=bheb_ml_... URL) to load all saved scans and re-scan commands.