Documentation
What Evidova does, how to send your conversations in, and what every score means.
Evidova scores conversations your AI agents already had. You send them in, we return a verdict on each one with the turns behind it. Nothing about your stack changes.
Start here
QuickstartConnect one platform and read your first result. Ten minutes, no code.Find your platform12 platforms with a page each, and what we read from every one.Use the APIPost conversations from your own stack, and read the scores back out.
Everything here
Start
How it works
How it worksThe six stages a conversation passes through, from arriving to being scored.Why an outside judgeWhy the thing scoring your agent should not be the thing that built it.Agreement with peopleHow we measure whether our scores match what your reviewers would have said.What gets scoredWhy the fast rules run on every conversation and the ones that cost money run on a share.What a score isWhat a number here means, what it does not mean, and what you must not use it for.
Guides
Find why people gave upRead the reasons customers stopped, ranked by how many conversations each one cost.Prove a fix workedShow that a change moved the number, over the same time range and the same traffic.Check you can trust the scoresRead the agreement figures for your own workspace before you act on a score.Investigate one conversationOpen a single conversation and see which turn broke which rule, and why.Add a rule of your ownWrite a question you want asked of every conversation, and put it live.Approve a suggested ruleReview a rule we drafted from your own documents before it counts.Set an alertGet told when quality or latency drifts, without being told forty times a day.Send a report to a clientProduce a dated report you can forward to a client, and the conversation list behind it.Add an agent or a clientRegister another agent, or another customer whose traffic you keep separate.Give someone one client onlyLet a partner see their own conversations and nothing else.Retention and redactionChoose how long transcripts live, and strip personal details as they arrive.Watch conversations as they landFollow traffic in real time and catch a bad release while it is happening.Change the words on screenWhat you can rename today, what the product calls things, and where the words come from.
API
APISend conversations in, and read the scores back out. Two surfaces, one key format.Authentication and API keysThe two scopes, what each one reaches, and how to replace a key.Webhook endpointThe one URL a platform posts finished conversations to.Send eventsPOST finished conversations from a stack of your own.Backfill historyUpload an export, so a score has history behind it from the first day.ConnectionsWhat a connection carries, why an API key cannot create one, and the two settings that decide whether it gets scored.Event payload referenceThe event shape to send: required fields, accepted fields, and what happens to the rest.ConversationsList conversations, and open one with its turns, findings and scores.Scores and findingsThe number, and which rule broke on which turn. Two endpoints, one reader's job.MetricsBucketed readings, so you can chart quality beside your own numbers.Errors and status codesEvery status both surfaces return, and what to do about each one.