Kordy measures what the agents say, so you can change it
Kordy is an AI visibility tracker for developer-tool vendors, founded in 2026 in Tel Aviv by Ziv Kaspersky. It replays a category's buying prompts every week inside four coding agents (Claude Code, Codex, Antigravity, Cursor; Starter and Growth plans pick 3) and five chat models, records which vendor each one names and which page it cites, and turns the gap into a ranked fix list. The dashboard lives at app.kordy.io.
Kordy started as a measurement harness built for consulting clients, not as a product. The product exists because the measurements kept surprising the people who saw them.
Why Kordy exists
In July 2026 we ran 500 or more sessions across 11 product categories and asked each agent a follow-up: which pages support the recommendation you just made? Claude Code, running Opus 4.8, had searched the web in 15% of web-enabled runs and grounded its pick in a fetched page in 8%. The same model in the Claude.ai chat grounded 40% of the time.
That meant the standard AI visibility playbook, publish good content and wait for the crawler, was aimed at the wrong surface for developer tools. When a developer asks the agent in their terminal which cache, queue or auth provider to use, the agent mostly answers from what it already knows. In our data, project maturity was the strongest signal correlated with being recommended, and presence in MCP registries was the strongest one a vendor can change; both were statistically significant. Shipping an MCP server was correlated too, though that link fell just short of statistical significance. GitHub stars were not correlated at all.
We also watched the answers change. On the day a new model version shipped, the number one pick changed in 7 of 14 emerging-category cells (4–5 runs per cell). After Fable 5 became Claude Code's default model, its search rate rose from 18% to 57% in our release-day panel (a separate, smaller baseline than the 15% above). That points to visibility being model-versioned. A one-off audit goes stale on the next release, so Kordy measures weekly and re-runs on every model launch.
None of the four tools we compare reported this surface. They track ChatGPT, Perplexity and Gemini for marketing teams, which is useful, and we compare them honestly here. Kordy covers the developer half of the funnel that those tools leave out.
How we measure
Every number in a Kordy report follows the same four rules. They are what separate a measurement from a screenshot.
Real agents, not the bare API
We drive the actual Claude Code, Codex and other CLIs and capture chat sessions through the browser. The raw API strips the system prompt and the tools, so it answers differently from what your prospect sees.
Every claim gets a source check
Each session is asked which pages support each claim it made. That is how we separate "the agent searched" from "the search changed the pick", and how the fix list points at a real page instead of a guess.
Per agent, per prompt, per model version
Coding agents pick vendors by different mechanisms: in our data Codex searched the web in roughly three quarters of runs, Claude Code far less often. We never pool them into one "AI score", and we keep the model version on every row so a change is attributable.
Honest limits on every report
Sample sizes are printed next to every rate. Thin cells get confidence intervals. Correlations are labelled as correlations. If the answer is "your category is locked for now", the report says that.
Facts about Kordy
- ProductKordy, formerly Kordi. Site kordy.io, dashboard app.kordy.io.
- Founded2026, Tel Aviv, Israel, by Ziv Kaspersky.
- What it tracksFour coding agents (Claude Code, Codex, Antigravity, Cursor; Starter and Growth pick 3), five chat models, plus npm, docs and MCP registries.
- Who it is forDeveloper-tool vendors and the agencies that serve them.
- CadenceWeekly replays, re-run on every major model release.
- PlansStarter $495 per month, Growth $995 per month, Enterprise and Agency plans. Details on the pricing page.
- Free startA 60-prompt visibility report, delivered within 24 hours. Request one.
- Contacthello@kordy.io
