Read the chat. Then reply well.

WHAT IT READS

Four signals, each as a probability.

Intent

Is the other person more likely romantic, friendly, merely polite, or creating distance? You get the whole distribution, not a single verdict.

Momentum

Is the conversation growing, steady, uncertain, or fading over the messages you pasted?

Reciprocity

Who is putting in the effort: balanced, you lead, they lead, or both low?

Next-step readiness

Does the chat support an invitation, more conversation, a clarifying question, or some space?

These questions are fixed and each answer lists every option with its probability. That is the kind of bounded judgment Jev is built for; see what Jev is for the model itself.

EXAMPLE CASE · ILLUSTRATIVE

A friendly chat that could go either way.

Maya writes: “I had fun. We should do this again sometime.” You answer: “Me too. Coffee Thursday?” She replies: “Thursday works :)”. The example below is illustrative, not a real user result.

Because romantic interest is not the leading estimate, the drafts stay relaxed and low-pressure instead of escalating. With a different distribution the drafts change.

HOW IT WORKS

Two steps, one BeatAPI account.

1. JEV estimates. It reads your pasted chat and returns probabilities for the four signals.
2. DeepSeek drafts. A language model receives the chat and those odds and writes three replies: steady, light, and direct. Jev does not write text; the drafting is a separate model call.
3. You decide. Copy a draft, edit it, or send nothing.

The tool lives on BeatAPI. Sign in there, top up your balance there, and each run uses that same balance for both steps. This site never runs the analysis, holds an account, or takes payment. Developers who want the same primitive in their own product can create an API key on BeatAPI and read the API guide.

LIMITS

Odds, not mind reading.

• Probabilities are model estimates from the text you paste, not facts about what someone feels.
• Tone, history, sarcasm, and context outside the chat are invisible to it.
• English works best; other languages are supported but less reliably. Test on conversations you already understand before trusting a pattern.
• Use it as one signal. Important relationship decisions stay yours.

PRIVACY

Analyze conversations you are part of.

Only paste chats you took part in, and remove details you would not want a processor to see. The analysis is handled by BeatAPI and its model providers under BeatAPI’s own terms and privacy policy, not on jevapi.io, which stores no chat text.