Lettertrace review (2026): a free, open-source AI visibility monitor you run on your own API keys
Lettertrace is MIT-licensed and free, hosted or self-hosted; you pay model providers per run. Its AI Overviews figure is a Gemini simulation. Who should use it, and when to move on.
Lettertrace is the rare AI visibility tool with no price card at all. The Letter Company publishes it under the MIT license, and you can use it free on lettertrace.com or self-host it on Next.js and Supabase through Docker or Vercel. Either way you bring your own API keys for Anthropic, OpenAI, Google and Perplexity and pay those providers directly for every run, with no markup.
For a procurement team, "free" raises its own questions: what does it actually measure, what will the API calls cost, and who maintains it? This review answers those from the GitHub README and the site, both read on 20 October 2026.
What it is and who it's for
You add topics. Lettertrace generates natural prompt variations and runs them against Claude, ChatGPT, Gemini, Google AI Overviews or Perplexity Sonar. It finds brand and competitor mentions by word-boundary matching, then a second LLM call classifies sentiment and whether you were recommended. Each run produces visibility, share of voice, prominence and sentiment, trended over time. With web search on, it records the sources cited, and you can map a prompt to a target URL to check whether your page is the one getting cited.
The fit is developers and technical marketers who want an inspectable tracker they can script. There's a REST v1 API, an MCP server, a CLI on npm (lettertrace) with OAuth 2.1 sign-in, a Docker image and team seats.
What it costs
| Option | Price | What you get |
|---|---|---|
| Hosted (lettertrace.com) | Free | Full app with your own API keys; you pay providers per run |
| Self-hosted | Free (MIT) | Next.js and Supabase via Docker or Vercel; your keys, your database |
The real cost is API usage, and that depends on how many prompts you run, on which models, and how often. The founder's Product Hunt launch post puts typical spend at "maybe $3". That is his estimate, not a measured figure, so run a week of your own schedule and check your provider bills before you rely on it.
What works
The engineering is careful, and the README is more candid about measurement than most paid tools. Gemini, Google AI Overviews and Perplexity need their own keys because, as the README explains, their search grounding doesn't survive a router. If you use an optional router key (Concentrate or OpenRouter), Lettertrace tests that web search is actually forced on when you save the key. That's a detail many trackers skip.
Self-hosting keeps the data in your own Supabase project, and API keys are encrypted with AES-256-GCM. Schedules can be daily, weekly or a custom interval through a cron endpoint. The README also says the hosted site runs the RB2B visitor-identification pixel on its public marketing pages only, and that self-hosted deployments ship without third-party tracking. Privacy teams will appreciate the disclosure.
What to check before you rely on it
- Your numbers are API answers. Lettertrace queries provider APIs, including Perplexity Sonar, not the ChatGPT, Claude, Gemini or Perplexity apps your customers use. Results can differ.
- Its "Google AI Overviews" isn't the real Overview. The README describes it as "a Gemini call plus a forced-search system prompt of ours". Treat that metric as an approximation.
- Self-hosting takes work: a Supabase project, an encryption key and, off Vercel, your own cron job.
- It's young. The GitHub repository was created on 21 July 2026, and on 20 October it showed 75 stars and 12 forks, with a push the day before. The CLI is at version 0.1.
Where it falls short of a full AI visibility platform
Lettertrace monitors mentions. It has no crawler logs, no traffic attribution and no content tools. You can see that Claude mentioned a competitor and cited their comparison page. You can't see whether ClaudeBot has ever fetched your own comparison page, or whether the people who click through from ChatGPT convert.
When Promptwatch is the better buy
Start with the free comparison. Promptwatch also has a free plan: Explore gives you 10 prompts on ChatGPT in one project, with no API bills. Promptwatch's data comes from real UI monitoring, not only APIs, so it reflects the answers people see in the apps, and Google AI Overviews and AI Mode are monitored directly rather than simulated.
When you need to connect mentions to outcomes, the paid plans add the parts Lettertrace leaves out. Agent Analytics shows which AI crawlers hit your pages, the crawl-to-citation path and crawl errors. Citation analytics breaks sources down by page and domain, down to Reddit and YouTube threads. Visitor analytics ties AI referrals to conversions. Essential at $95 a month includes 50 prompts on all tracked LLMs; Professional at $245 adds crawler logs (25M) and 150 prompts.
Developers keep programmatic access: Promptwatch has an MCP server and a REST API on every paid plan, plus a Claude Connector and Slack integration.
Verdict
Lettertrace is a well-built free option, and its README is a good read for anyone who wants to understand how AI visibility measurement can go wrong. Use it to get a cheap trend line on API answers, or as a scripting playground. Don't report its AI Overviews number as the real thing, and don't expect it to explain why you're cited. When you need that, Promptwatch's crawler logs and visitor analytics are the next step.
FAQ
Is Lettertrace really free?
Yes. The code is MIT licensed and the hosted version is free. You pay the AI providers directly for the API calls each run makes.
Which AI engines can Lettertrace monitor?
Claude, ChatGPT, Gemini, Perplexity Sonar and a Gemini-based approximation of Google AI Overviews, all through their APIs.
What's the best paid upgrade from Lettertrace?
Promptwatch, if you need real UI data, crawler logs and conversions. Its Explore plan is free too: promptwatch.com.