> ## Documentation Index
> Fetch the complete documentation index at: https://docs.giwa.tokscale.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Five Proven Patterns
> Comparables, market sizing, and the structural fixes that keep Tokscale out of the web3 reputation graveyard.
Every layer of Tokscale maps to a comparable that already worked, with named traction numbers and named sources. Estimates are flagged as estimates.
## The reference stack
| Tokscale layer | Proven by | Hard traction (2025 to 2026) | Tokscale adaptation |
| --------------------------------------- | ------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Multi-model gateway | OpenRouter | \~$1.3B valuation1; ~100T tokens/month (5x in 6 months); 8M+ developers; ~$50M ARR2 at a \~5% take rate | Same gateway, crypto-paid and verified, on GIWA |
| Per-request crypto payments | x402 (Coinbase) | 100M+ transactions on Base in under a year; Linux Foundation governance with Visa, Mastercard, Stripe, Google, KakaoPay | Real inference demand as x402's organic use case |
| Exchange-KYC attestation on own L2 | Coinbase Verifications on Base | \~9,300 KYC attestations on day one, 15% of all EAS attestations at the time | Upbit (13M+ users) to Dojang to Tokscale gating: the identical play |
| Builder identity and score | Talent Protocol (Base) | 1M+ Builder Passports minted; integrated by Base, Etherscan, ENS; Coinbase Ventures-backed | Dojang-bound Passport recording verified AI work, not public-data aggregation |
| Ecosystem bounties to verified builders | Superteam Earn (Solana) and Base Builder Rewards | \~\$10M+ distributed to \~198K builders; Base pays 20 ETH/week to top-100 builders ranked purely by verified reputation | GIWA-funded bounty board keyed to Builder Passports |
1Series B, May 2026.
2Sacra estimate, not a disclosed figure.
We add the one layer none of them has: an un-forgeable, exchange-KYC-bound record of how developers actually work with AI. OpenRouter's known weakness is that routing commoditizes as models converge; the identity layer is the fix. The gateway generates the data, and the Passport is the moat routing-only players cannot copy.
## B.AI: the proof, and the adaptation
B.AI is the closest crypto-native gateway reference. It publishes no verifiable traction metrics, so treat it as a directional analog, not a benchmark.
### How B.AI works
| Component | B.AI implementation |
| ----------------- | ------------------------------------------------------------------------------------------------------------------ |
| Gateway | OpenAI-compatible unified API to 20+ LLMs (GPT, Claude, Gemini, DeepSeek, MiniMax, Kimi) with intelligent routing |
| Billing | Credit-based: 1 USD = 1,000,000 credits; prepaid top-up |
| Payments | TRON-native tokens (TRX, USDT, USDD, USD1) plus fiat cards; x402 payment standard for agent autonomy |
| Identity | ERC-8004-based agent identity; on-chain address as permanent identifier |
| Positioning | Anonymity-first: no mandatory KYC |
| Adjacent products | BAIclaw browser extension; BAI Code, also usable as a Claude Code backend, the same integration path Tokscale uses |
| Plans | Pro \$200/mo and Max \$1,000/mo subscription tiers alongside pay-per-use |
What it proved: paying subscribers at $200/mo and $1,000/mo for crypto-paid AI inference. That is market recognition for the category, and GIWA has nothing in it.
### The adaptation, the GIWA way
| Axis | B.AI (TRON) | Tokscale (GIWA adaptation) |
| ------------------- | ----------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------- |
| Identity philosophy | Anonymity, no KYC | Verification via Dojang, because KYC is GIWA's whole thesis |
| Identity subject | AI agents (ERC-8004) | Human developers (Dojang-bound Passport); agents later as delegated actors |
| What accrues | Agent transaction history | Developer work history |
| Existing data | Starts at zero | Real usage data from Tokscale early adopters |
| Value of a record | ERC-8004 identity is an ERC-721 token, transferable by design, so a reputation can be bought and sold | Bound to a Dojang-verified human, non-transferable by construction |
**B.AI gives agents a wallet. Tokscale gives builders a verifiable work history.**
An anonymity-first gateway cannot run on GIWA, so the GIWA-native version of this category has to be verification-first. For more information, see [Why GIWA](/overview/why-giwa).
## The ERC-8004 transferability gap
ERC-8004 has been live on Ethereum mainnet since January 29, 2026, backed by the Ethereum Foundation, MetaMask, Google, and Coinbase, with 21,500+ agents registered across EVM chains within weeks. It provides Identity, Reputation, and Validation registries for agents.
But its identity token is ERC-721-based and therefore transferable: the owner can sell or delegate the agent identity, reputation included. For agent commerce this is tolerable. For work history it is fatal. A credential you can buy is not a credential.
Tokscale's answer is to anchor identity to a Dojang Verified Address (human, KYC-linked off-chain, attested via EAS) and treat usage proofs as attestations to that identity, not tokens owned by it. We stay ERC-8004-aware, with agents as delegated actors under a verified human operator in the Phase 3 roadmap, without inheriting its weakness.
## Adjacent categories
These are the comparisons a reviewer reaches for first, answered directly.
| Category | Examples (traction) | Why Tokscale is different |
| --------------------- | ------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| Model routers | OpenRouter (\~100T tokens/mo, \~\$1.3B); LiteLLM (46K+ GitHub stars); Vercel and Cloudflare AI gateways | No crypto payments, no identity, no verification; routing alone is commoditizing |
| Agent payment infra | B.AI, Nevermined, Skyfire | Agent-centric, anonymity- or billing-focused; no human work history |
| Generic web3 identity | ENS (2.8M+ names, 850+ integrations); Gitcoin and Human Passport (2M+ users); Galxe (14M+ users) | Broad identity, no proprietary data source; Tokscale owns the gateway that generates its proofs |
| Builder reputation | Talent Protocol (1M+ passports) | Aggregates public GitHub and on-chain data anyone can recompute; Tokscale's usage data is generated inside its own paid product |
| Usage trackers | Local token counters (ccusage-style) | Self-reported, forgeable, no economic layer; nobody has monetized verified usage |
We do not pitch a generic identity layer head-on. We pitch one vertical: verifiable AI work history, with generic identity as long-term vision only.
## Market sizing
### TAM: inference spend and agentic payments
The AI inference market is forecast to grow from $106B in 2025 to $255B in 2030, a 19.2% CAGR (MarketsandMarkets). Enterprise LLM API spend hit $8.4B by mid-2025, roughly doubling annually (Menlo Ventures). Agentic commerce forecasts range from $1.5T (Juniper) to $3T or $5T (McKinsey) by 2030. A gateway taking OpenRouter's \~5% of inference flow addresses a multi-billion-dollar fee pool.
### SAM: crypto-paid inference plus verified identity for APAC builders
The Asia-Pacific LLM market reaches \~$94B by 2030. Korea has 16M+ crypto investors, about 32% of the population, and ranks second globally in on-ramp volume (Chainalysis). The AI-native developer population is expanding fast: 84% of developers use or plan to use AI tools and 51% use them daily (Stack Overflow 2025); Claude Code went from $0 to \$2.5B ARR in about nine months; coding is 51% of all enterprise generative-AI usage (Menlo Ventures).
### SOM: the GIWA ecosystem, years one and two
These are targets, not claims. GASOK cohorts plus GIWA's early builder base, with Upbit's 13M+ users as the distribution funnel (Dunamu 2025 operating revenue \~₩1.56T). Benchmark trajectories: Talent Protocol reached \~200K passports within a year of a single-chain launch, and Superteam Earn reached \~\$10M distributed on an ecosystem-funded model. Year-one target: 500 to 1,000 Passports and the first won of verified inference spend.
### The credential demand signal
The AI-skill wage premium hit 56% in 2025, up from 25% (PwC Global AI Jobs Barometer). Over one-third of entry-level jobs now require AI skills, nearly triple fall 2025 (NACE). Only 29% of developers trust AI output (Stack Overflow). Employers need to screen how people work with AI, and no verification layer exists. That is the Builder Passport's market.
Number hygiene: OpenRouter's \~\$50M ARR is a Sacra estimate, some AI-tool ARR figures are analyst estimates, and agentic-commerce TAMs vary by definition. Estimates are labeled as estimates wherever they appear.
## The graveyard: why web3 reputation dies, and why Tokscale survives
Most on-chain reputation projects failed. The diagnosis, and the point-by-point structural fix:
| Failure mode | Evidence | Tokscale's structural fix |
| ---------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------- |
| No proprietary data source | POAP (7.2M badges, maintenance mode as of March 2026), DegenScore, generic scores: all aggregated public data anyone could recompute | Usage data is generated inside our own paid gateway: proprietary, un-forgeable, impossible to reconstruct externally |
| No economic loop | Scores that unlock nothing get ignored | Earn layer (real bounties), x402 payments, and paid verified-builder search give the Passport cash value |
| Transferable credentials leak signal | POAP tradability killed its meaning; ERC-8004 identity is a sellable ERC-721 | Non-transferable Dojang-bound EAS attestations |
| Token-first, product-second | \$TALENT collapsed to a \~\$60K market cap: a reputation token is not a business | Revenue is the gateway take rate on real inference; no token in the MVP, matching GIWA's tokenless, compliance-first design |
| Speculative volume is not organic demand | Artemis and CoinDesk analysis (March 2026): much of x402's headline volume was wash or memecoin churn, with \~\$28K real daily volume and wallet retention falling from 87% to 5% | Every Tokscale payment maps 1:1 to a real inference request: organic by construction |
The last row doubles as offense: x402 has Linux Foundation governance, card-network membership, and 100M+ transactions of rails, but a shortage of genuine non-speculative demand. A gateway where developers pay for actual inference is precisely the organic workload the protocol needs, and GIWA can host it first.
## Next steps
* For why these comparables land on GIWA specifically, see [Why GIWA](/overview/why-giwa).
* For the layer none of the comparables has, see [Builder Passport](/builder-passport/overview).
* For what Tokscale has actually shipped against this thesis, see [Live Evidence](/overview/live-evidence).