Claude Fable 5.1 · Claude Code
A model wrote all of this.
Hyperion listens to meetings. RevOps closes the books. Quote Desk prices the deal. Three working tools, written end to end by Claude Fable 5.1, and this site along with them.
Hyperion, live on a meeting
It hears what the room misses.
Participants open one link. The browser listens. Four AI subsystems run at their own cadence and post only what deserves a glance: the disagreement nobody caught, the fact nobody looked up, the decision that is still open.
- MayaSo we ship the pricing page Friday. Everyone good?
- TomSure. I mean, once legal signs off on the new terms.
- MayaRight, that's separate though. Friday for the page.
- PriyaWhat was our conversion on the old page again?
- TomSomething like two percent? I'd have to check.
- MayaFine. Let's move on to the launch email.
Maya heard “yes, Friday.” Tom said “yes, after legal.” Those are different dates, and nobody has noticed.
Old pricing page conversion, last quarter: 2.4% (from the analytics export shared in this room).
The real decision: ship the page without the new terms, or wait for legal. It has not been made.
Three chapters
Each tool lives where its users already work: a meeting, a monthly close, the minutes after a sales call. None of them is a demo. The previews below reproduce the products faithfully with synthetic data, because the real ones carry real revenue and real prices.
Photo: CoWomen on Unsplash
Anyone in a meeting who wants to see the conversation while they are in it
A shared web page that listens to a meeting and surfaces what is being missed: misunderstandings, unstated assumptions, drift, and the question nobody looked up.
Photo: Jordan Encarnacao on Unsplash
People can only process one stream, what is being said. Hyperion runs the second stream, the meta-conversation, and shows it on a canvas every participant can glance at without leaving the discussion. There are no fixed card types; the model decides what deserves attention.
Hyperion's feed on a fictional meeting: transcript on the left, what the model surfaced on the right.
- Stack
- Next.js 16, React 19, Tailwind v4, Vercel
- Transcription
- Deepgram Nova 3 streaming, word timestamps, diarization
- Models
- Claude Sonnet 5 with web search, Gemini 3.1 Pro, plus Crusoe and NVIDIA providers behind one interface
- Real-time
- Ably pub-sub with token auth per room
- Observability
- Neon Postgres call log, Datadog RUM and LLM Obs
Photo: Jonathan Velasquez
Operators who need a revenue number they can trust
One dashboard that unifies customers, subscriptions, renewals, and revenue across Stripe, AWS Marketplace, GCP Marketplace, and direct invoices, anchored to the general ledger. A Stadil product, white-labeled by Scalr.
Photo: Mika Baumeister on Unsplash
Four billing rails each have their own dashboard and none of them shows the company. RevOps merges them into one entity per customer, keeps committed and on-demand revenue as two streams that are never silently combined, and renders any difference from the books as an explicit reconciliation gap rather than a plug.
RevOps with illustrative figures and fictional customers. The real instance runs on the deploying company's books.
- Stack
- Python 3.12 pipeline, Next.js 16, React 19, Tailwind, TypeScript
- Sources
- Stripe, AWS Marketplace, GCP Marketplace, QuickBooks Online, Mixpanel, Ramp
- Views
- Revenue, customers, accounts, receivables, retention, growth, forecast, cash, spend, KPIs, financials
- Storage
- One committed data file per build; no server, no runtime API calls
- Close cycle
- Monthly; closed months anchor to the GL, open month from billing sources
Photo: Denny Müller
Sales teams, in the minutes after a discovery call
A static single-page app that prices a quote from the company's real discount curves and prints it as a branded, Letter-size PDF the buyer can forward to procurement. A Stadil product, white-labeled by Scalr.
Photo: Marcus dePaula on Unsplash
It replaces two pricing spreadsheets. The pricing model lives in one pure module with tests that pin known outputs, so a quote is priced the same way twice. The form is a deal ledger that serves a live paper preview: what you see is what the customer receives. There is no backend, so there is nothing to host or log into.
Quote Desk with a synthetic quote. The ledger on the left prices the paper on the right; the real curves stay private.
- Stack
- Vanilla ES modules, one HTML file, no build step required
- Pricing
- One pure module, tests pin spreadsheet outputs
- Paper
- US Letter, measured pagination, print-dialog PDF
- State
- Round-trips through the URL hash; copy link is the share mechanism
- Distribution
- Single-file build, hosted as a private artifact
Photo: Fabian Kleiser
Range
Each cell names the concrete instance in that product. A blank means the product had no need for it.
| Skill | Hyperion | RevOps | Quote Desk |
|---|---|---|---|
| Real-time systems | Mic to Deepgram WebSocket, fragments fanned out over Ably, four AI cadences | – | Live paper re-renders on every keystroke of the ledger |
| Multi-model orchestration | One provider interface over Anthropic, Gemini, Crusoe, NVIDIA; model per route is a one-line edit | – | – |
| Financial correctness | Per-call cost computed at write time from a rate table | GL-anchored recognition, deferred waterfalls, explicit reconciliation gaps | Discount curves and tiers pinned by tests against the spreadsheet |
| Data pipelines | Circleback transcripts synced into a parser-checked corpus | Six sources pulled, merged by curated identity maps, built into one file | – |
| Print and document layout | – | Close-package schedules in the accountant's own format | US Letter paper paginated by measurement, rubric conventions |
| Evaluation and observability | Prompt lab, eval harness, text-to-SQL ask box over the call log | Per-build review files for names, invoices, and books differences | Headless Chrome print check of the PDF |
| Design systems | Type scale, warm nebula, one authored motion per surface | Instrument-panel black, IBM Plex, fixed channel hues | Two materials never mixed: dark chrome, white paper |
| Documentation discipline | Developer guide updated in the same PR as the change it describes | Truth lives in code and working notes; README states only long-term facts | Dated design spec archived, product and design docs kept current |
Photo: Egor Myznik on Unsplash
About the model
The newest model in Anthropic's Claude 5 family and the most capable Claude model generally available. It sits in the Mythos-class tier above Claude Opus and shares its underlying model with Claude Mythos 5.1, with additional safety measures for dual-use capabilities.
Everything on this site came out of Claude Code sessions: the pricing engine and its tests, the revenue pipeline and its reconciliation rules, the meeting assistant and its prompt lab, each product's design system and documentation, the shader in the first viewport, and these words.