Skip to content

PRICING.md

Latest commit

 

History

History
153 lines (121 loc) · 6.93 KB

PRICING.md

File metadata and controls

153 lines (121 loc) · 6.93 KB
Raw

Pricing decision note — price the autonomy layer

2026-07-20, written under the full-tilt mandate ("business model may be adjusted… document the change in a PRICING.md decision note"). Status: DECIDED (copy re-anchor shipped in this PR) + QUEUED (enforcement follow-up).

Addendum 2026-07-21 — COGS lock-in (Nate-directed)

The only real inference COGS is a chat turn (~$0.01–0.06). Watchers don't burn inference: the guardian is a per-minute deterministic cron, DCA/jobs compile without the model. So the exposure was the credit allowances, and they're now sized so a maxed plan can never exceed its price in COGS:

Plan Was Now Grandfathered
Builder $0 2,500 250/mo (+40/day cap) n/a (no Stripe object)
Growth $99 25,000 8,000/mo pre-2026-07-21 subs keep 25k forever
Scale $499 150,000 40,000/mo pre-2026-07-21 subs keep 150k forever

Enforced in lib/plans.ts planCreditsFor (keyed on subscription createdAt). Stripe untouched — allowances aren't Stripe objects.

Circuit breakers (lib/billing.ts — the "leave it open" guarantee):

  • FREE_DAILY_TURN_CAP (default 40/day/wallet, free tier only)
  • HOUSE_DAILY_TURN_CAP (default 2,000/day across EVERYONE ≈ $60/day worst-case Anthropic bill; env-overridable, clamped) Both refuse with honest per-gate copy; standing jobs/DCA/guardian are never gated (they don't touch the model). Fail-open on store errors (chat availability beats metering, unchanged).

Queued for Nate on Stripe (owner step): credit packs — $10 per 1,000 credits as a one-time price, so heavy chatters buy margin-priced inference instead of hitting walls. Code hook lands when the product id exists.

B2C stays subscription-free: DCA/jobs/guardian price via the 0.20% flow fee (now on LiFi + Uniswap v3 + CoW, website#485) — a $100/week DCA pays $0.20/week, no seat fee. Subscriptions are for embed hosts and power autonomy (standing-intent capacity 3/25/unlimited).

If a faster watcher ever ships (1s–10s cadence): it's a compute SKU ($5–10/mo per protection or Scale-only), never inference-priced — the tick loop must stay model-free.

The decision

Keep the three plans and their prices — Builder $0 / Growth $99 / Scale $499, the 0.20% flow fee, and credits as the meter for attended chat. Do NOT touch the live Stripe products. What changes is what the tiers are FOR: the differentiator moves from embed-site count + credit volume to standing-intent capacity — how much of your money moves without you watching.

Plan Standing intents (active jobs + DCA schedules + guardian protections)
Builder $0 3
Growth $99 25
Scale $499 Unlimited

Everything else (sites, credits, theming, orgs, SLA) stays as-is per tier.

Why

  • The strategic frame: the company is a take-rate on money that moves unattended. Standing intents are the retention engine; one-shot attended swaps are a commodity. Seats and credits price the commodity; capacity for standing intents prices the thing that recurs.
  • Three free standing intents is deliberately generous enough for the aha (a DCA + a guardian stop + one job) and small enough that a working portfolio of automations is a paid behavior.
  • The falsifiable link: the attended/standing scoreboard (website#478) shows whether standing money concentrates in accounts that would hit these caps. If the standing line grows and never touches a cap, the caps are wrong — revisit with data, not vibes.

What ships when

  1. Now (this PR): /pricing copy re-anchors on the autonomy layer (lib/plans.ts highlights). Advertising a limit before enforcing it harms nobody — free users get MORE than advertised until enforcement lands.
  2. Follow-up card (not today): soft enforcement at creation time — the 4th active standing intent on Builder gets a friendly upgrade chip, not a wall; nothing running ever pauses for plan reasons. Kill switches and safety gates are NEVER plan-gated.
  3. Never: retro-limiting. Grandfathering rule: anything active at enforcement time stays active forever; caps gate NEW creations only; existing subscribers keep whichever terms are better.

Explicitly not changed

  • Stripe products/prices (no new objects, nobody re-checkouts, nobody strands — the handoff's hard rule).
  • The 0.20% flow fee and its single source (lib/fees.ts).
  • x402 paid doors: per-call packaging for agent customers, separate lane.
  • Credits: still the attended-chat meter on every plan.

Addendum 2026-07-21 — Intent links: the creator rail (Nate-approved)

Intent links (website#500/#503) get the third capacity axis and the first revenue-share program. Decided in-session with Nate; shipped ledgered (phase 1) in website#505's lane.

The fee map (unchanged rule, now on the record)

Charge on conversions, never on movements or inflows. The fee is earned when Yeetful's routing chose a price for you — one asset became another through our venue cascade:

Action Fee
Swaps, tokenized-stock buys/sells, DCA runs 0.20% (lib/fees.ts, unchanged)
NFT sales / listings / transfers $0 forever (a sale is an inflow)
Sends, bridges, funding legs $0 (taxing the fix kills the wedge)
Reads, asks, votes, staking $0

Creator fee-split

  • 50% of the fee (10bps of notional) accrues to the link's creator on fee-bearing conversions attributed to their link. Single source: lib/fees.ts CREATOR_FEE_SPLIT + FEE_BEARING_BUILD_PATHS (exactly the four native-swap-* paths — NFTs/transfers/bridges move $ through links but never earn).
  • Sybil-proof by construction: creators earn a fraction of fees actually paid; self-referral is a self-discount, not a drain.
  • Phase 1 (live): ledgered. Fees land in the treasury unchanged; earnings compute read-time from embed_turns (guardrail-priced, server side); claims open at $10, paid manually as USDC on Base.
  • Phase 2 (when payouts get frequent): venue fee recipient becomes a per-creator deterministic split contract (0xSplits on Base/Eth/Arb; CREATE2-derived so the guard re-pins it; 4663 stays ledgered). No user ever signs an extra transaction in either phase.
  • Disclosed on every creator-minted /i page: "The creator of this link earns half of Yeetful's 0.20% conversion fee."

Active-link capacity (mirrors standing intents)

Builder $0 Growth $99 Scale $499
Active intent links 3 25 Unlimited

Soft at mint time (402 + upgrade pointer; revoke frees capacity); revoked links keep their funnel + earnings history forever; nothing retro-limits. Link visitors' auto-run turns bill the creator's credit allowance — the embed-key contract in link form; the existing breakers bound the worst case.

Explicitly not changed (again)

  • Stripe products/prices; credit allowances + breakers (#486 lock).
  • The global money-moved metric stays guardrail-priced embed_turns — per-link client-reported funnel numbers never feed it.