An open-source (MIT) AI visibility monitoring platform from The Letter Company. You define your brand and buyer-relevant topics, and it generates natural question variations, queries Claude, ChatGPT, Gemini, Perplexity, and Google AI Overviews using your own API keys, then records how often your brand and competitors are mentioned, recommended, sentiment-scored, and what share of voice each holds. It is BYOK throughout—Lettertrace itself charges nothing, and you pay providers directly with no markup. The application can be self-hosted via Docker or Vercel on your own Supabase, keeping data in your infrastructure, or used on the hosted free tier at lettertrace.com. Developer surfaces include a REST API, an MCP Server, and a CLI (npm install -g lettertrace) with OAuth 2.1 and --json machine-readable output. Two caveats are documented by the project itself: Gemini and Perplexity require their own keys because search grounding fails on aggregated routers, and “Google AI Overviews” is simulated via a Gemini call with a forced-search system prompt rather than being real Google AI Overviews data. The project is very young—the GitHub repository was created on 21 July 2026, the CLI is at version 0.1, and it had roughly 75 stars by late October 2026.
Pros & Cons:
Pros: BYOK with no platform markup means the cost of monitoring is provider API pricing rather than a subscription, which suits teams already holding keys; MIT-licensed and self-hostable on your own Supabase, so brand and competitor data can stay in your infrastructure; covers five AI surfaces (Claude, ChatGPT, Gemini, Perplexity, and a simulated Google AI Overviews) rather than a single engine; records mentions, recommendations, sentiment, and share of voice together, which is more actionable than a raw citation count; developer-native with REST API, MCP Server, CLI, OAuth 2.1, and JSON output; the hosted tier is free, lowering the barrier to try it.
Cons: Extremely early-stage—repository created July 2026, CLI at version 0.1, modest star count, and limited maintainer bandwidth, so stability and longevity are unproven; “Google AI Overviews” is simulated, not real Google data—README states it is a Gemini call with a forced-search system prompt, and this must not be marketed as genuine AI Overviews monitoring; a static security scan reported undisclosed egress—the documentation claims no network access while the source contains numerous network calls and a child_process execution surface, which warrants independent review before production use; self-hosting requires maintaining Supabase and scheduled jobs, so it is not a one-click deployment; the hosted free tier routes data through Lettertrace’s servers, which may not suit brands with sensitive competitive data.
Who is it for?:Technical marketing and SEO teams tracking how AI assistants describe their brand relative to competitors; agencies running AEO/GEO programmes that need share-of-voice and sentiment data across multiple engines; developers who prefer BYOK and self-hosting over a closed SaaS subscription; teams already using MCP-capable tools that want AI visibility data inside their existing workflows.