Loading article…

Loading 4weird…

DocsGame AI & Buddy

Docs · your co-pilot

A coach who watches you play.

Dialogue bots, AI directors, voice acting - plus a universal 9-voice Gaming Buddy that reads the screen and coaches you live. All metered with the 25% cut included.

9: voices
0.5-2x: speech speed
2: AI kinds: built-in + Buddy
1 widget: every play page

Know what you're playing with

Required vs. optional AI

Detail and play pages wear badges disclosing mode + provider + '25% cut included'. Learn to read them:

Required AI

The core loop needs it - e.g. an attack director driving enemies. Playing the game is using AI, metered per turn, decision, or GPU minute.

Optional AI

Toggleable dialogue bots, directors, TTS. Off means zero AI metering for that feature. Mix freely with the Buddy.

The voice cast

Nine voices, one default diva

Nova ships as the default; all nine ride the same meters (chat tokens + voice characters + optional screen snapshot + database writes, 25% inside).

Alloy

Ash

Coral

Echo

Fable

Onyx

Nova

DEFAULT

Sage

Shimmer

How a session sounds

Buddy 101

gaming buddy - live session

9.2¢ this session

Buddy · Nova 1.2x

Nice dodge! Two pickups spawn left in ~5s - grab the shield first 🛡️

coach my aim for this boss?

session 4.1¢total 38.7¢24h 12.0¢1h 4.1¢

Open any play page, start a Buddy session, ask for what you want: “coach my aim,” “explain this level,” “call out pickups.”
Keep sessions scoped - one game, one goal. Shorter sessions cost less and coach better.
End the session when you stop playing - open sessions keep the door open for metered turns.
Guests: Buddy, multiplayer, saves, and game AI are signed-in only. Sign up to unlock them.

Under the hood

A Buddy turn: observe → reason → act → meter

Every chat turn reuses the VibeCodeWorker loop shape. The widget observes the screen, the server reasons with a real model (or a clearly-labelled free fallback), acts by speaking, and meters the true cost - tokens + image + database - with the 25% cut inside.

POST /api/buddy/chat - one turn

true-cost metered

# observe: what the widget forwards

{ "game_slug": "gravegain2d", "screen_text": "…≤2000 chars",

"score": 1200, "voice": "nova", "session_id": "…",

"message": "coach my aim?", "history": […], "brain": "auto" }

# reason → act → meter: reply + per-turn cost breakdown

✔ reply + cost { chat, image, db }

👀 Observe is explicit, never sneaky. The client forwards game slug, title, screen text, score, and optionally a downscaled JPEG/PNG snapshot plus an opt-in camera frame. Every distinct frame reaches the model for that turn only - never stored, never logged - and each bills as image input tokens on the chat leg.
🧠 Reason picks its brain by key. brain: "auto" (default) prefers the OpenAI Responses API when its key is set, falls back to OpenRouter chat-completions, and returns a clearly-labelled local reply at zero cost with neither. You can pin "openai" or "openrouter" explicitly - or send "stream": true for live token deltas over SSE instead of one JSON blob.
💬 History is short and amnesiac by default. Up to 8 recent turns ride along for context and are never stored server-side; cross-session memory only exists when you opt in (see below). Typed questions cap at 500 chars, screen context at 3000 in the final prompt.
🚦 Chat and voice throttle separately (tens of requests per minute per player) so a rapid-fire coaching session degrades into "slow down" instead of a surprise bill. Game-AI metering behind it caps quantities per call for the same reason.

Shortcuts

Slash commands + the intent router

Don't type an essay mid-boss-fight. The widget expands shortcuts into full model-ready prompts, and a keyword router steers the persona before the model even sees the turn.

/tactics …

One concise tactical recommendation for the current battle state, focused on whatever follows the slash. The action catalog's coaching button fires this for you.

/hail …

A short fictional enemy-commander radio taunt, then one fair counter-tactic. Roleplay first, coaching always.

/react …

React to the moment in 1-2 short sentences - the default lens when you just want company, not counsel.

/vibe …

Match energy from the opt-in camera frame: warm, kind, never diagnosing, never identifying.

Continuity

Sessions group spend, memory is opt-in

Sessions are the wallet's unit of grouping: start one when you sit down, end it when you stand up, and the widget plus /my/usage/ show live session spend. Memory - the Buddy remembering you across sessions - is a separate, explicit yes.

Start, resume, end
Open with POST /api/buddy/session { action: "start" } (game slug + voice), close with { action: "end" }. Reloaded mid-game? GET /api/buddy/session?open=1 lists your open sessions newest-first so the widget resumes instead of orphaning a meter.
Opt into memory per game
Memory lives per game behind GET/POST /api/buddy/memory. POST without consent: true stores nothing, full stop. What's stored is an extractive rolling buffer of plain "Player: … / Buddy: …" lines - zero model calls, zero AI spend - merged and trimmed oldest-first, then carried into future prompts as one short capped block.
Presence costs extra, on purpose
The optional 3D avatar meters by the minute and camera check-ins by the frame through POST /api/buddy/presence, each with a sane per-call cap so a buggy loop can't drain a wallet in one call. Off by default; the coach works fine without either.

Volume knob

Voice: models, speed, and the free fallback

Voice output is metered separately from chat - and only when provider voice actually speaks. Everything else is free by construction.

🎛️ Two models, nine voices, quarter-step speed. POST /api/buddy/tts takes up to 2000 chars with tts-1 (default) or tts-1-hd, any of the 9 voices, and speed clamped to 0.5-2x in 0.25 steps. HD costs more per character - the per-turn breakdown says exactly how much.
🔇 No key, no problem, no bill. Without a provider key the route returns a fallback flag and the widget speaks through your browser's built-in speech - free, instant, slightly more robotic. Provider hiccup mid-session? Same fallback, same zero charge.
🎨 Game voices have their own meter. Built-in game TTS/NPC lines meter through POST /api/game-ai/meter by kind (dialogue, director, tts, runpod-gpu, inference, buddy-*) with the same 25% cut - toggle the feature off and that meter stops dead.

Price tags

What each leg costs

Buddy turns bill true upstream cost converted to gross Vibe Coins (provider dollars × 100 ÷ 0.75, 25% inside, centicentcoin resolution, minimum a hair above zero per leg). Built-in game AI meters at fixed kind rates instead - same cut, simpler math:

💬 Dialogue

3 / 1k tokens

NPC chat, quests, Buddy conversation

🎬 Director

2 / decision

Spawns, pacing, difficulty calls

🔊 Game TTS

2 / 1k chars

In-game voice lines, announcers

🎙️ Buddy voice

per-char meter

tts-1 or tts-1-hd + DB leg, true cost

🖼️ Avatar

0.08 / min

Optional 3D presence, off by default

📷 Camera

0.03 / frame

Opt-in emotion/body-language frames

The lineup

Which games ship AI today

Games declare AI features in a static registry the play shell reads without a database round-trip - so the 25% badge renders instantly. Everything listed today is optional: toggle it off, pay nothing for it.

🪦 GraveGain 2D + 3D

Optional NPC dialogue bot plus an optional RunPod-backed game director pacing spawns and encounters.

🧟 Last Words: Zombies

Optional director tuning zombie horde waves to your typing speed, plus zombie voice-line taunts in any of the 9 voices.

🐾 Assassin Animals

Optional director adapting facility layouts and patrols to your stealth, via a hosted inference endpoint.

🦈 BattleSharks 2 + more

Optional mutation announcer and arena directors (RunPod GPU), each badged with mode + provider + cut on the detail page.

Loading article…

DocsGame AI & Buddy

Docs · your co-pilot

A coach who watches you play.

Dialogue bots, AI directors, voice acting - plus a universal 9-voice Gaming Buddy that reads the screen and coaches you live. All metered with the 25% cut included.

9: voices
0.5-2x: speech speed
2: AI kinds: built-in + Buddy
1 widget: every play page

Know what you're playing with

Required vs. optional AI

Detail and play pages wear badges disclosing mode + provider + '25% cut included'. Learn to read them:

Required AI

The core loop needs it - e.g. an attack director driving enemies. Playing the game is using AI, metered per turn, decision, or GPU minute.

Optional AI

Toggleable dialogue bots, directors, TTS. Off means zero AI metering for that feature. Mix freely with the Buddy.

The voice cast

Nine voices, one default diva

Nova ships as the default; all nine ride the same meters (chat tokens + voice characters + optional screen snapshot + database writes, 25% inside).

Alloy

Ash

Coral

Echo

Fable

Onyx

Nova

DEFAULT

Sage

Shimmer

How a session sounds

Buddy 101

gaming buddy - live session

9.2¢ this session

Buddy · Nova 1.2x

Nice dodge! Two pickups spawn left in ~5s - grab the shield first 🛡️

coach my aim for this boss?

session 4.1¢total 38.7¢24h 12.0¢1h 4.1¢

Open any play page, start a Buddy session, ask for what you want: “coach my aim,” “explain this level,” “call out pickups.”
Keep sessions scoped - one game, one goal. Shorter sessions cost less and coach better.
End the session when you stop playing - open sessions keep the door open for metered turns.
Guests: Buddy, multiplayer, saves, and game AI are signed-in only. Sign up to unlock them.

Under the hood

A Buddy turn: observe → reason → act → meter

POST /api/buddy/chat - one turn

true-cost metered

# observe: what the widget forwards

{ "game_slug": "gravegain2d", "screen_text": "…≤2000 chars",

"score": 1200, "voice": "nova", "session_id": "…",

"message": "coach my aim?", "history": […], "brain": "auto" }

# reason → act → meter: reply + per-turn cost breakdown

✔ reply + cost { chat, image, db }

👀 Observe is explicit, never sneaky. The client forwards game slug, title, screen text, score, and optionally a downscaled JPEG/PNG snapshot plus an opt-in camera frame. Every distinct frame reaches the model for that turn only - never stored, never logged - and each bills as image input tokens on the chat leg.
🧠 Reason picks its brain by key. brain: "auto" (default) prefers the OpenAI Responses API when its key is set, falls back to OpenRouter chat-completions, and returns a clearly-labelled local reply at zero cost with neither. You can pin "openai" or "openrouter" explicitly - or send "stream": true for live token deltas over SSE instead of one JSON blob.
💬 History is short and amnesiac by default. Up to 8 recent turns ride along for context and are never stored server-side; cross-session memory only exists when you opt in (see below). Typed questions cap at 500 chars, screen context at 3000 in the final prompt.
🚦 Chat and voice throttle separately (tens of requests per minute per player) so a rapid-fire coaching session degrades into "slow down" instead of a surprise bill. Game-AI metering behind it caps quantities per call for the same reason.

Shortcuts

Slash commands + the intent router

Don't type an essay mid-boss-fight. The widget expands shortcuts into full model-ready prompts, and a keyword router steers the persona before the model even sees the turn.

/tactics …

One concise tactical recommendation for the current battle state, focused on whatever follows the slash. The action catalog's coaching button fires this for you.

/hail …

A short fictional enemy-commander radio taunt, then one fair counter-tactic. Roleplay first, coaching always.

/react …

React to the moment in 1-2 short sentences - the default lens when you just want company, not counsel.

/vibe …

Match energy from the opt-in camera frame: warm, kind, never diagnosing, never identifying.

Continuity

Sessions group spend, memory is opt-in

Start, resume, end
Open with POST /api/buddy/session { action: "start" } (game slug + voice), close with { action: "end" }. Reloaded mid-game? GET /api/buddy/session?open=1 lists your open sessions newest-first so the widget resumes instead of orphaning a meter.
Opt into memory per game
Memory lives per game behind GET/POST /api/buddy/memory. POST without consent: true stores nothing, full stop. What's stored is an extractive rolling buffer of plain "Player: … / Buddy: …" lines - zero model calls, zero AI spend - merged and trimmed oldest-first, then carried into future prompts as one short capped block.
Presence costs extra, on purpose
The optional 3D avatar meters by the minute and camera check-ins by the frame through POST /api/buddy/presence, each with a sane per-call cap so a buggy loop can't drain a wallet in one call. Off by default; the coach works fine without either.

Volume knob

Voice: models, speed, and the free fallback

Voice output is metered separately from chat - and only when provider voice actually speaks. Everything else is free by construction.

🎛️ Two models, nine voices, quarter-step speed. POST /api/buddy/tts takes up to 2000 chars with tts-1 (default) or tts-1-hd, any of the 9 voices, and speed clamped to 0.5-2x in 0.25 steps. HD costs more per character - the per-turn breakdown says exactly how much.
🔇 No key, no problem, no bill. Without a provider key the route returns a fallback flag and the widget speaks through your browser's built-in speech - free, instant, slightly more robotic. Provider hiccup mid-session? Same fallback, same zero charge.
🎨 Game voices have their own meter. Built-in game TTS/NPC lines meter through POST /api/game-ai/meter by kind (dialogue, director, tts, runpod-gpu, inference, buddy-*) with the same 25% cut - toggle the feature off and that meter stops dead.

Price tags

What each leg costs

💬 Dialogue

3 / 1k tokens

NPC chat, quests, Buddy conversation

🎬 Director

2 / decision

Spawns, pacing, difficulty calls

🔊 Game TTS

2 / 1k chars

In-game voice lines, announcers

🎙️ Buddy voice

per-char meter

tts-1 or tts-1-hd + DB leg, true cost

🖼️ Avatar

0.08 / min

Optional 3D presence, off by default

📷 Camera

0.03 / frame

Opt-in emotion/body-language frames

The lineup

Which games ship AI today

🪦 GraveGain 2D + 3D

Optional NPC dialogue bot plus an optional RunPod-backed game director pacing spawns and encounters.

🧟 Last Words: Zombies

Optional director tuning zombie horde waves to your typing speed, plus zombie voice-line taunts in any of the 9 voices.

🐾 Assassin Animals

Optional director adapting facility layouts and patrols to your stealth, via a hosted inference endpoint.

🦈 BattleSharks 2 + more

Optional mutation announcer and arena directors (RunPod GPU), each badged with mode + provider + cut on the detail page.

A coach who watches you play.

Required vs. optional AI

Nine voices, one default diva

Buddy 101

A Buddy turn: observe → reason → act → meter

Slash commands + the intent router

Sessions group spend, memory is opt-in

Start, resume, end

Opt into memory per game

Presence costs extra, on purpose

Voice: models, speed, and the free fallback

What each leg costs

Which games ship AI today

A coach who watches you play.

Required vs. optional AI

Nine voices, one default diva

Buddy 101

A Buddy turn: observe → reason → act → meter

Slash commands + the intent router

Sessions group spend, memory is opt-in

Start, resume, end

Opt into memory per game

Presence costs extra, on purpose

Voice: models, speed, and the free fallback

What each leg costs

Which games ship AI today