What it does
Four things, in a bar you can read at a glance.
Jungler tracker
Last-seen time, an elapsed gank timer and live KDA, with warning and danger states.
Objective timers
Dragon, Baron, Void Grubs, Rift Herald and Scuttle Crab, with the rules checked against each patch.
13:32 Dragon soul secured
Match Scout
Press F10 for rank, last-8 form and win rate on the champion each player locked in. Labels are small-sample heuristics.
38%Player C Gold IVWEAK
Champ select picks
Three AI suggestions. Click a card to swap it for a fresh pick; dismissed picks never come back.
Also: a resizable Dashboard window with an alert feed, a post-game review and a battle log.
Decision 1 of 4
Information, not instructions.
My reading of Riot's developer policy: no real-time overlay that tells players what to do right now. So version 2.0.0 reworded every live alert and the live AI prompt. The overlay states the situation and the player decides.
Pre-game and post-game advice were already fine. Reflective coaching is allowed.
Wording examples are illustrative of the change, redrawn as mock-ups.
Decision 2 of 4
Don't let the AI guess what it can look up.
An early build told players to build a nonexistent item path. Two fixes followed.
- Real recipesBuild paths for 170+ items come from Riot's Data Dragon feed, cached and refreshed every patch.
- A freshness filterTips are dropped if they quote a countdown of 5 seconds or less, or name a champion's life status that live data contradicts.
- Spoken, or silently droppedA late or wrong tip never reaches the player.
Decision 3 of 4
Know when you're out of your depth.
The app once gave Summoner's Rift advice, such as Void Grubs and Dragon, in ARAM matches. The cause: game mode was checked once, on the first tick. It is now checked every tick from several signals.
When League Classic arrived with different rules, the app learned to say so and stop, instead of being confidently wrong.
Unsupported mode (name shown here) — coaching disabled
Decision 4 of 4
Say when it didn't work.
- The matchup research line says "loaded" or "unavailable, using general knowledge", so players know if a tip is grounded.
- Match Scout labels are marked as small-sample heuristics (last 8 games), not rigorous statistics.
- AI and network errors give a plain reason, such as a certificate problem from a firewall, instead of a blank panel.
Release record
Shipped in the open.
- v1.9.5 to v1.9.7ARAM fixes across alerts, game plans and AI coaching.
- v1.9.8Disable tracking in unsupported game modes.
- v2.0.0Rebrand to Rift SensAI, Match Scout, and the Riot compliance pass.
- v2.0.2Fixed the Gemini error for new users, found through a GitHub report.
- v2.1.0Dashboard window and jungler gank timer, inspired by a contributor's pull request.
- v2.1.1Stability pass: socket timeouts, atomic config saves, URL protocol checks.
- v2.2.0Champ select swap and Gemini 3.5 Flash-Lite.
Dates from the public GitHub releases page.
Under the hood, and the limits
Plain tools, honest edges.
Details and what is not done
- The game state is polled every 2 seconds. No latency figures are claimed.
- Hotkeys use a low-level keyboard hook, because the game bypasses standard global shortcuts.
- The app does not modify game files, inject code or automate gameplay. It is not endorsed by or affiliated with Riot Games.
- Not shipped yet: an Overwolf-based build, which would let the overlay and hotkeys work without leaving the game.