Paint alongside Jev on a shared emoji canvas

01 · Role in the system

What Jev does here

Every stroke triggers one Jev request. The choice set is typed and situational: emojis that share keywords with what is painted around the stroke, each offered with a named placement - above the line, at its end, inside the loop - so the model decides both what to add and where. A second yes/no asks whether the stroke is unfinished, letting Jev close a loop or carry the line on when it is sure. The pick is drawn from its probabilities, wider the less confident, and the top candidates flash on the canvas before the chosen emoji lands. Only drawing data leaves the browser - stroke shape, endpoints, neighbouring emoji, a small text picture of nearby cells - never who painted. Calls go through Vercel's AI Gateway with TypeSafe direct as fallback, per-visitor rate limits and a daily site spend cap. Everyone's strokes arrive live; ink returns five cells a second.

02 · Control boundary

Where Jev sits

Jev as a co-creator on a live shared canvas: one typed call per stroke chooses an emoji and a named placement from situation-built options, a yes/no gates continuation, and gateway routing with spend caps keeps the toy affordable.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Emoji options are keyword-matched around each stroke; the model cannot invent motifs outside the emoji set.
• Live site depends on gateway availability; TypeSafe direct is only the fallback path.
• Brand-new project (0 stars at review time); no external evaluation.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• cwdx ↗Community · github · public · checked 2026-09-29