Saylent

The open-source audit of what AI assistants say about your brand

With the receipts: every verdict traced to the answer, the cited page, and the fix. Run it on your own domain, with your own keys.

npx saylent demo

Renders a full sample report on your machine first: no key, no network, $0.

npx saylent audit your-domain.com --dry-run

Then price a real audit before anything is sent. One OpenAI or Anthropic key runs it with that engine. Two keys turn on the cross-family judge; Gemini and Perplexity keys add those engines.

All four: about $0.80 to $1.20 a run, of your own credits. Our two recorded runs cost $0.93 and $1.11.

Open the full sample report: a real run against Kestrel Uptime, a fictional company we audit.
Eight screens of the Saylent app in sequence: your brands, the question editor with its live price, the summary, the full report, the fix tracker, the rival comparison, the operator's keys and models, and the operator's budget and kill switch
Open the live demo: the same run inside the app, with history, a fix tracker and share links. Nothing saves, no account.

Check what AI crawlers can read on your site

Type a domain. In a few seconds you see which AI crawlers your site lets in: robots.txt, a live per-crawler fetch, your schema and your meta directives. No keys, no sign-up, no LLM calls, nothing stored.

This reads your gates, not your answers. It cannot tell you whether ChatGPT, Claude, Gemini or Perplexity actually recommend you, who they recommend instead, or which sources they cite. That needs the full audit.

Run it once

One command, your own keys, and a report you can open from disk, email, or drop in a ticket. It prints what it is about to spend before it spends anything, and a dry run costs $0. Ask ChatGPT, Claude, Gemini and Perplexity, add Google AI Overviews, AI Mode and Copilot with your own SerpApi key, import the questions your buyers really search, and point it at your server logs to see which AI crawlers actually read which pages and how many visitors the assistants sent.

Keys, the command, the report

A terminal running npx saylent audit: the engines, judge and profile lines, then all nine numbered stages, then the verdict, the band and the report path

What the command hands you: four receipts, one screenshot each

The verdict block of a report: the headline finding, recommended and mentioned counts, and the top rival

Verdict

"Kestrel Uptime is in none of the 11 scored answers. Upcheck is named most, 17 times, and upcheck.example is the page they cite most."

See it in the sample report

Per engine, the pages its answers cited and how often

Receipt

The exact answer, the engine, the date, and the page it was built from.

See it in the sample report

The site access checks: one row per AI bot, with the status and what it means

Gate

Which AI bots your site lets in, tested live, not guessed from robots.txt.

See it in the sample report

A drafted fix, with the quoted answers that justify it

Fix

Drafted, evidence-anchored, ready to ship: JSON-LD, a comparison page, entity copy.

See it in the sample report

Every screenshot is generated from the sample run, never hand-edited.

Keep score as a team

Two people, five brands and a year of runs do not fit in a folder. Deploy the same engine with a memory: history, a fix tracker that knows what you shipped, scheduled verifies, share links, and alerts to Slack, email or a signed webhook when an answer changes. One deployment, as many user accounts as you like, each isolated from the others, on infrastructure you own.

Every screen of the app

The Saylent dashboard: a next-move strip, a brand card with its recommended score and buttons to open the latest report or run a verify, and a recent-runs table

Operate it for others

Somebody owns the keys, the bill and the blast radius. The operator console is the daily spend cap, the kill switch that stops every run at once, a model per role with no redeploy, and an append-only log of who changed what.

The operator console

The budget and limits screen: a seven-day spend table against the daily cap, a Pause all runs kill switch, and a daily spend cap field with a required reason

Automate

The same checks on every push, inside your agent, or inside your own program. The CI gate costs $0 and needs no keys; the Claude Code plugin, the MCP server and the library spend only your own credits, under the same ceiling the command respects. Every table exports as CSV or JSON, and the app has a read-only API with personal tokens.

The Action, the Claude Code plugin, the MCP server, the library

.github/workflows/ai-access.yml

on: [push]
jobs:
  ai-access:
    runs-on: ubuntu-latest
    steps:
      - uses: yotambraun/saylent@v0
        with:
          domain: example.com

Understand

Nine stages, each one narrating itself and each one saying where you can change it. Presence is a deterministic alias match, everything else is judged by a model from the other provider family, and the report states a band instead of a number the draws cannot support. Each answer keeps the web searches the engine ran before writing it, where the engine reports them. Every cited page is counted, so the report shows which sites shape the answers and which ones name your rivals and not you: the off-site moves with the most receipts behind them.

How it works, stage by stage

The nine audit stages in order: crawl, brand model, questions, engines, judge, cited pages, site gates, fix plan, score

Project

Apache-2.0, solo-maintained, and the test suite runs at $0 with no keys and no database. Adding an answer engine is one file implementing one interface; adding a site check is one registry entry and one test.

Contribute, or read the changelog

What it costs

Real numbers by profile, from recorded runs, before you run anything. Your keys, your dollars.

What gets sent to providers

Exactly what leaves your machine, to whom, and what never does. No telemetry, anywhere.

Compared with hosted tools

Sourced prices, dated, and the five things the hosted tools still do that this does not.

Run it on your own domain

npx saylent audit your-domain.com --dry-run

Open the live demo · Star on GitHub