This page separates what is live and verifiable today from projections and vision. Figures labelled as projections are estimates, not results. Where a claim depends on information we have not independently confirmed, it is marked
TODO.Live today (verifiable)
On Base mainnet
Escrow contract
0xf808b264E13Bf809C8e86afaF4e14c200931101E, verified on Basescan. USDC settlement is in production.Working product
@CanvasVerificationBot runs the full loop: join interception → conversational task → Kimi scoring → USDC payout. Advertiser and group-owner dashboards are live.Whitelist phase
Groups are manually approved during early access to establish quality baselines before the permissionless marketplace opens.
Documented end to end
Architecture, scoring, smart contract, and anti-gaming are all documented — see Architecture.
The problem
Every messaging platform and much of the open web runs the same toll booth — a bot-check on entry — hundreds of millions of times a day. Cloudflare has estimated humanity spends about 500 years per day solving CAPTCHAs. That attention is real, recurring, and captured today by the platforms running the checks. Three parties lose:Owners earn nothing
Tens of thousands of joins hit communities daily. None of that activity pays the person who built the community.
Advertisers can't reach them
The most engaged audiences on the internet live inside chat platforms, where per-verified-human reach doesn’t exist.
Data buyers pay for noise
AI labs pay for volume from context-free labor pools when what they need is verified, context-matched humans.
The solution
Canvas replaces the bot-check with a sponsored task at that exact moment, then settles payouts on Base once Kimi scores the response as genuine. The captcha itself becomes the ad channel — and the data channel. Full mechanics: How it works.- A user requests to join a registered group and is muted and DM’d a short task
- The Canvas agent runs a 2–4 message conversation generated from the advertiser’s brief and the group’s topic
- Kimi scores the transcript; a passing score accrues a USDC payout, split 90% to the owner and 10% to Canvas, settled on Base
Illustrative example (hypothetical). A perps protocol sponsors a task in a Base DeFi group: “What can you do on this protocol beyond perps — vaults, staking, or points?” The user answers, Kimi scores it genuine, the owner is paid, and the protocol can offer the verified user a direct chat with its own agent. Verification becomes the first product touchpoint, not just a gate.
Why now
Onchain USDC settlement, cheap Base transactions, and LLM scoring good enough to grade open-ended conversation only recently became simultaneously available. Canvas needs all three: programmatic per-verification micropayments, and a scorer that can tell a genuine human answer from a pattern response. The incumbents (Google, Cloudflare) will not build this — pricing verified human attention onchain means giving up the data-and-attention moat they capture for free.Differentiation
The captcha is the channel
Reach a context-matched human at the one moment they’ve self-selected into a specific community — not an impression, a verified interaction.
Provenance you can't fake later
Each completion carries the group context, a Kimi quality score, and a verifiable on-chain settlement transaction. This can’t be reconstructed by scraping audience lists after the fact.
The bigger market: verified human data
Crypto is the wedge because both sides already speak the language. The market is every fandom and interest community. The same mechanism that verifies a DeFi trader can collect a preference label from a verified fan — reasoning data that exists nowhere else, because it requires a verified member explaining their own taste, in their own words, at the moment they chose to join. This is the RLHF thesis behind Canvas: context-matched, quality-scored human completions with delivered provenance.Economics
Canvas takes a 10% protocol fee; the group owner keeps 90% of the winning bid. The floor price is $0.10 per verification, and advertisers bid above it — niche, high-intent groups command real competition, undifferentiated ones settle near the floor. Phase 2 re-verification turns a group’s existing membership into recurring revenue at the same rate. Worked figures: Business model and Payouts.Our June 2026 deck modelled a 0.10 floor and a 90/10 split** (owner/Canvas), matching the deployed contract and docs.
TODO: Founder input required — confirm which model is canonical going forward and align the deck.Market size (projections)
TAM — RLHF platforms
Global RLHF platform market projected at 111.1B (2034). Source: Grand View Research, per company deck (June 2026).
Closest comp — Telegram verification
Rose, the leading Telegram verification bot, is cited at ~2M monthly active chats — the distribution Canvas targets. Source: Rose bot FAQ, per company deck (May 2026).
Traction
Canvas is in mainnet beta, whitelist phase. The contract, bot, scoring, payouts, and dashboards are live and verifiable.Go-to-market
Supply — group owners
Direct outreach to crypto KOLs running active Telegram communities. The first owner earning real USDC from an existing group becomes the case study; supply grows from PR, not a sales motion.
Demand — advertisers
Crypto protocols and agents already understand verified attention — no education required. This is the deliberate, revenue-bearing half of the market.
Architecture
A Node/TypeScript agent server on Railway orchestrates Telegram, a conversational LLM, the Kimi scorer, Postgres state, and the Base escrow contract. Strict TypeScript, a test suite, and a Foundry-tested contract back it. Full detail: System overview, Verification flow, Scoring, Smart contract.Team
Rohit Sarkar — Product & Infrastructure
Years across crypto infrastructure (0x, Caldera, InfStones). Selected for The Vault (an EigenCloud × ETH Foundation builder residency) as one of five featured projects, where the original Canvas architecture was built.
Alexander Vicol — Product & Engineering
Full-stack systems engineer specializing in Base-native infrastructure and production onchain AI agents. Base Batches 002 alumnus; shipped agent infrastructure featured in the Base ecosystem.
Roadmap
Phase 1 (live) → Phase 2 re-verification (recurring revenue) → Phase 3 permissionless marketplace (staking, anti-sybil, self-serve) → Phase 4 Solana. Detail and targets: Roadmap.Try it
Add the bot to a group
Start a Telegram group with
@CanvasVerificationBot, make it an admin, and you’re set. Questions or partnership: hello@canvasprotocol.xyz.