> ## Documentation Index > Fetch the complete documentation index at: https://docs.signalmd.dev/llms.txt > Use this file to discover all available pages before exploring further. # Introduction > SignalMD tells the owner of a production voice agent what to fix and whether the fix worked. SignalMD answers one question for someone who owns a voice agent in production: what should I fix, and did my fix work? You connect a Retell agent. SignalMD reads its calls, finds the problems that repeat across them, and orders them by how much they hurt your callers. After you ship a change, it shows what production did before and after, so you can see whether the change helped or the numbers only moved. SignalMD is for the product manager, QA lead or AI engineer who runs a Retell voice agent and can edit its prompts, flow and tools. Retell is the only platform SignalMD supports at launch. ## What you get Connect an agent. SignalMD imports your latest calls (up to 500 from the last 30 days) and drafts a short [brief](/concepts/the-brief) of what it is for. You approve the brief, and every call is then read against it. From then on SignalMD keeps up with new calls on its own. You get: * **Issues with a place, a cause and a change.** Not "dead air", but "a lookup at one step takes nine seconds", with a suggested fix and quotes from the calls behind it. Confirmed defects sit in **Fix**, and changes you control that move your numbers sit in **Improve**. When the calls do not show why something happens, SignalMD says **Cause not established** and does not guess. See [Issues](/concepts/issues). * **Version comparisons.** When you publish a change in Retell, SignalMD shows what changed and what production did afterwards, and tells you when a difference is real and when there are not enough calls yet. See [Versions](/concepts/versions). * **Tests from real calls.** A call behind an issue can become a test: a simulated caller with the original caller's goal, run against a version of your agent. Testing is opt-in and needs a separate Retell key that can write, plus a monthly budget you set. See [Test a fix](/guides/test-a-fix). New workspaces open on a read-only demo agent built from synthetic calls, so you can look around before you connect anything. ## How it differs from reading transcripts A transcript shows you one call. SignalMD reads every call and groups what it finds by the change that would fix it, so three symptoms of one slow tool become one row. It makes its own reading of each call and does not rely on Retell's summary, sentiment or success flag. See [How SignalMD works](/concepts/how-signal-works). ## Limits to know about * **New calls show up within about 15 minutes.** If you paste SignalMD's webhook URL into the agent in Retell, each call arrives as soon as it ends. Analysis takes a few minutes more. See [Connect Retell](/guides/connect-retell#how-signal-gets-data-from-retell). * **Confirmed patterns only.** **Fix** and **Improve** list problems SignalMD has confirmed across several calls, with a cause it established. A confirmed pattern without an established cause is listed apart, under **Cause not established**. Anything not yet confirmed is kept and appears once it is. * **English works best.** SignalMD reads calls in other languages, but its analysis is tuned for English. ## Where to go next Sign up, connect a Retell agent and read your first issues. What SignalMD gives you, from your first calls to a fix that worked. Fix an issue, compare versions and test a fix. A chat beside every page that reads your issues and drafts fixes. Read issues, calls and versions from your own tools and CI. Give an AI assistant read access to the same data. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.