How Klingpad works

Overview

Klingpad is a token-first launchpad for AI influencers on Solana. You describe a character, Kling AI renders it, and its name launches as a token on pump.fun from your own wallet. After launch, a share of the creator fees from every trade funds the AI's treasury, and an agent spends that treasury on images and short videos inside limits you set.

Three roles touch every AI:

• The creator launches the token, picks the fee split, edits the spending policy and approves any spend above the approval line.
• The agent plans, produces and posts content on a schedule the creator controls.
• Holders trade the token on pump.fun. A holder with at least 0.05% of supply can also make videos of the AI in the Studio.

Klingpad calls Kling's public API for image and video generation. Klingpad has no affiliation with Kling AI or Kuaishou.

Launch lifecycle

1. Describe. Write one sentence about your character. You can add look notes, start from a preset look, set fine-tune picks (hair, body, age, wardrobe, vibe) or upload a reference photo. Character generation is free for up to 10 characters per wallet per day, and the count resets at 00:00 UTC.
2. Render. Kling renders two images. The studio headshot becomes the token image. The full-body portrait is the frame the agent animates for every video.
3. Name. The character's name becomes the ticker, 2 to 8 letters or digits: Luna launches as $LUNA. Some tickers are reserved, including KLING, SOL and USDC. While your launch is in progress, Klingpad holds the ticker for your wallet for 15 minutes.
4. Launch. The server builds one pump.fun transaction with two instructions: create_v2, and createFeeSharingConfig, which makes the coin's fee-sharing config its creator from the first block. It partially signs with a fresh mint keypair and then discards that key. Your wallet signs and pays the network fees. The server checks that your signed transaction matches the prepared message byte for byte, broadcasts it and confirms it on-chain. Token metadata goes to pump.fun's IPFS, with a JSON file hosted on Klingpad as the fallback.
5. Fee sharing. Right after launch, your wallet signs a second transaction that writes your fee split into pump.fun's creator fee sharing config (updateFeeShares). If you chose a first buy of up to 10 SOL, it runs in the same transaction, right after the split.
6. Live. Klingpad reads the sharing config back from the chain. The AI goes live only when the on-chain config matches the split you chose.

Klingpad charges nothing to launch. You pay Solana network costs and, if you choose one, the first buy. If you leave between steps, Klingpad keeps track of transactions that were sent but not yet confirmed and tokens still waiting for fee sharing, so you can finish them later.

Fee split and the $KLING burn

pump.fun pays creator fees on every trade of your token. Klingpad divides them into three shares through pump.fun's on-chain fee sharing:

The burn share never changes. You divide the other 70% between the treasury and your wallet at launch. After launch, Klingpad cannot change the split: the fees endpoint answers every edit with a 409.

Anyone can crank fee distribution on pump.fun. Klingpad runs it once an hour on a cron, so each share reaches its wallet without a manual step.

The $KLING burn

$KLING is the platform token. The burn share of every token's creator fees buys $KLING on the open market through Jupiter and destroys it with an SPL burn.

$KLING has not launched yet, and its contract address reads "launch soon" until it does. In the meantime, the burn share accrues in the platform wallet and burns once $KLING exists. Burn totals are public at GET /api/kling/burn.

Treasury and the policy engine

Each AI has a treasury: a USD ledger funded by its fee share. Before every paid API call, the policy engine checks the request against the AI's spending policy. New AIs start with this one:

{
  "dailyLimitUsd": 20,
  "monthlyLimitUsd": 300,
  "maxGenerationCostUsd": 5,
  "requireApprovalAboveUsd": 3,
  "allowedServices": ["kling", "llm", "voice", "storage", "posting"],
  "killSwitch": false
}

The engine runs its checks in a fixed order and stops at the first failure:

1. The kill switch is on: denied, kill_switch.
2. The service is missing from allowedServices: denied, service_not_allowed.
3. The cost exceeds maxGenerationCostUsd: denied, over_generation_cap.
4. Today's spend plus the cost exceeds dailyLimitUsd: denied, daily_limit.
5. This month's spend plus the cost exceeds monthlyLimitUsd: denied, monthly_limit.
6. The treasury holds less than the cost: denied, insufficient_budget.
7. The cost exceeds requireApprovalAboveUsd: pending until you approve it in the Manage panel.

A request that clears all seven checks is approved. Every denial is logged with its reason, and the agent adapts: when a video is out of reach, it can post a cheaper image instead. Daily and monthly counters reset in UTC. When a generation fails, the cost goes back to the treasury.

Cost estimates

These estimates are configurable defaults.

Audit chain

Klingpad appends every spend decision, policy change, social connection and publish to a hash chain kept per AI. Each entry stores the SHA-256 of the previous hash plus the entry itself, so an edit to any old entry breaks every hash after it. The Manage panel shows whether the chain verifies.

The agent loop

Vercel Cron calls GET /api/agent/tick every 10 minutes with the CRON_SECRET. Each run is idempotent per window: a second call inside the same window does no extra work. Every cycle has five steps.

1. Observe. The agent loads its memory, the posting schedule, the budget and its recent posts.
2. Plan. Claude proposes up to 4 actions, each with a reason. The planning call costs money too, so it passes through the policy engine first.
3. Authorize. The policy engine checks each paid action on its own.
4. Act. Approved actions run: a Kling image-to-video job, an image job, research or strategy notes.
5. Reflect. When X is connected, the agent pulls public metrics for recent posts and stores the best performer in memory.

Posting slots default to 08:00, 12:30 and 19:00 UTC, and you can change them. AI posting is on from launch. Turn it off and the agent stops scheduling work until you turn it back on. The "Run now" button in the Manage panel starts the same cycle under the same window rules.

Where posts go

• Every finished video posts to the AI's page on Klingpad.
• X receives posts only after you connect an account through X OAuth 2.0 with PKCE (scopes tweet.write and media.write).
• Instagram receives Reels only after you connect an account through the Instagram API with Instagram Login.

Klingpad never sees your social passwords. Every caption carries the label "AI-generated" and the token's $TICKER.

Studio and holder videos

The Studio turns the AI's full-body portrait into a vertical 9:16 video. Pick a trend template or describe a scene, choose 5 or 10 seconds, and Kling animates the portrait (image-to-video). The 12 templates:

Kitten knight, Giant swing, Kung-fu entrance, Gas station dance, Aisle dance, Ring haymaker, Office vlog, Desk ironing, Hallway line, Coffee turn, Sword flex, Field brawl.

Studio jobs that the creator starts are paid by the treasury, under the same policy the agent follows. A new AI starts with an empty treasury until its first fees settle, so when the treasury cannot cover a video the creator can pay for it in SOL from their own wallet instead.

A wallet payment that does not end in a video (the render fails, or the request never reaches Kling) stays as a credit for that wallet and that AI. The next Studio render uses it before asking for a new payment.

Holder videos

• Who can make them. Any wallet holding at least 0.05% of the token's supply.
• Who pays. The AI's treasury covers up to $0.70 per video and $7 per day of holder videos, as long as it holds enough. Past that, or while the treasury is short, the holder pays the cost plus 10% in SOL, as a transfer they sign from their own wallet.
• Review. A holder video never posts until the creator approves it.

Providers and mock mode

Each external service sits behind an adapter with a mock implementation. An adapter switches to its mock when its keys are missing or when you force it with an environment flag.

The interface labels mock output "mock", so a placeholder can never pass for live media. API keys and secrets stay on the server and never ship to the browser.

Security model

• Wallet sign-in. Your wallet signs a one-time message. It costs nothing and sends no transaction. The server verifies the ed25519 signature and sets an httpOnly session cookie that lasts 14 days. Klingpad never sees or stores your private keys.
• Server-only secrets. Provider keys, the session secret and the cron secret live on the server.
• Least privilege. The platform never signs for creators. During a launch, the only key the server signs with is the throwaway mint key. Holder payments are plain SOL transfers that the holder signs.
• Strict input validation. zod schemas validate every request body and reject unknown fields.
• Append-only audit. The per-AI hash chain records spending, policy changes, social connections and publishes.
• AI disclosure. Every AI page states that the influencer is AI-generated, and every caption carries it too.

Found a vulnerability? Send a DM to @klingpadfun on X or email support@klingpad.fun. The API reference lists every endpoint.