Get LLM credits,
at a discount.
One key, every model, no subscriptions.
SPARK is prepaid inference, sold on an open book below face value. Stake AGGR and every model gets cheaper still. Activate it, point your SDK at us, keep your code.
Pay in USDC or SOL · estimated from the book
Your request goes to the provider as sent. No rerouting, no cheaper stand-ins, no prompt rewriting.
Same model id, same weights, same output you would get going direct.
Prompts and completions are never written to disk. We keep billing metadata only.
One API key. Same models. Less spend.
OpenAI-compatible endpoint in front of 400+ models. Swap the base URL, keep the request, pay less for the same tokens.
const client = new OpenAI({ baseURL: "https://aggregat.xyz/api/v1", apiKey: "sk-aggr-…",}); await client.chat.completions.create({ model: "anthropic/claude-sonnet-4.5", messages: [{ role: "user", content: "Ship it." }],});At a 30% book level. 5% market fee applies at checkout.
Credit, priced by the market.
Holders list SPARK at a discount to face value. Buyers take the cheapest levels first. Settlement is on Solana.
Every model, one key.
Claude, GPT, Gemini, Llama, DeepSeek, Mistral and the long tail. List price on the left, what you pay with book credit on the right.
Every token, accounted for.
Per-key spend, latency and model mix. Cost comes back on every response, so your dashboard and your invoice agree.
- anthropic/claude-sonnet-4.5$189.60
- openai/gpt-5$98.92
- google/gemini-2.5-pro$61.83
- deepseek/deepseek-chat$37.10
- meta-llama/llama-4-maverick$24.73
The whole migration is two lines.
Base URL and key. Streaming, tools, JSON mode and vision work as they did. Every response carries its cost in a header.
import OpenAI from "openai"; const client = new OpenAI({- baseURL: "https://openrouter.ai/api/v1",- apiKey: process.env.OPENROUTER_API_KEY,+ baseURL: "https://aggregat.xyz/api/v1",+ apiKey: process.env.AGGREGATE_API_KEY, // sk-aggr-… }); // model ids stay the same: "anthropic/claude-sonnet-4.5"Prepaid inference. Cheaper for stakers.
SPARK is how you pay. Staked AGGR is how you pay less: the vault's USDC yield funds the compute that covers your discount.
- 01Buy SPARK
1 SPARK = $1 of usage at list price on Claude, GPT, Gemini and 400+ more. Pay USDC, take the book below face value, activate into API balance for any sk-aggr-… key.
- 02Stake AGGR, pay less
Tiers by share of supply: Bronze 0.01% → 10% off, Silver 0.05% → 20% off, Gold 0.1% → 30% off, Platinum 0.5% → 40% off. On every model, every request.
- 03Yield splits 80/20
The vault is a holder, so StonkFun pays it USDC. 80% tops up the compute reserve that covers discounts, 20% runs ops.
- 04Provider tokens add more
Hold the top AGGR-paired coin of a model family: +10% at 0.1% of its supply, +20% at 1% of its supply on that family's models. Capped at 60% in total.
Yield in, discounts out.
No subscriptions and no emissions. The USDC that AGGR pays its holders buys the compute behind every staker discount.
- Staked AGGRUSDC holder rewards on every token in the vault: the 3% tax, converted and paid by StonkFun
- Treasury holdingsUSDC holder rewards on the AGGR the treasury holds (dev buy at launch)
- 5% market feeon SPARK purchases from the book
- 10% of standard launchesof the creator fee on every standard launch here, buys AGGR and burns it
- 80%Compute reserve
OpenRouter balance. Covers the staker discount on every request, across every model.
- 20%Ops 10% + 10%
Marketing wallet and team wallet. Both public, so the split is verifiable on-chain.
No fixed buyback. The compute reserve targets 30 to 90 days of gateway spend; everything above 90 days buys AGGR and burns it. The more the gateway is used, the more of the yield goes to discounts; the less, the more goes to burns.
- Bronze · 0.01% of supply−10%
- Silver · 0.05% of supply−20%
- Gold · 0.1% of supply−30%
- Platinum · 0.5% of supply−40%
Over the 7-day compute budget, discounts scale down (floor 25%) and recover on their own.
The keeper measures OpenRouter balance in days of trailing 7-day gateway spend. Above 90 days, the excess buys AGGR and burns it. Ops is never touched, and nothing under $25 executes.
Every coin launched here trades against AGGR, so launchpad volume is AGGR demand. Reward coins pay their holders in AGGR. Standard launches send 10% of the creator fee to an AGGR buy and burn.
Shipping in the open.
- Token launch on StonkFun
- Stake AGGR → discount on every model
- OpenAI-compatible gateway
- SPARK order book
- aAGGR vault on-chain (custody + yield sweep)
- Launchpad paired with AGGR, provider tokens
- SPARK as an SPL token (activate = burn)
- DIEM-style perpetual credit (Venice model)soon
- Agent-native x402 payments
- Image, video and audio endpoints