Patent Log & 2026 Patentability Scan
Everything we drafted, everything that died, everything still worth filing. Internal / legal reference page.
TL;DR
- ~83% of the 18 claims are effectively dead — killed by either Alice v. CLS Bank (2014) (abstract-idea rejection) or by prior art now shipping in 2026.
- ~11% are narrow / contested — arguably novel in combination, but easy to design around and expensive to enforce.
- ~6% (one claim, partial) is still defensible — specific SSE tool-pill state machine +
_actiondispatch pattern narrowly claimed. - Better moat than any patent for a $9–$99 SaaS: trademark CASH.BOT ($350), copyright (automatic, free), trade secret on prompts/funnel, ship speed.
- Recommendation: file the trademark this week. Skip the utility patent. Provisional ($320 pro-se) only if "patent pending" on the landing page is worth a year of marketing flex.
The 18 Claims — Triage
DEAD — Alice §101, or § 102/103 prior art
WEAK — narrow / contested / cheap to design around
LIVE — still plausibly novel in 2026
Independent Claims (1–3)
DEAD
Claim 1. NL command → AI tool router → parallel microservice calls → SSE stream → data cards.
Killed by Alice (abstract "do business on a computer"). Covered by ChatGPT plugins (2023), LangChain agents, AutoGPT, Zapier AI, every function-calling demo since Claude/OpenAI shipped tool use. The "desktop metaphor" window-dressing doesn't survive §101.
DEAD
Claim 2. Browser-rendered desktop OS (windows/taskbar/start menu) hosting 62+ microservices.
Prior art since 2005: eyeOS, Silveos, Jolicloud, ZeroPC, CloudMe, glide.com. ChromeOS-in-a-browser is 10+ years old. "60+ services" is volume, not novelty.
DEAD
Claim 3. Tier-gated service access within the OS metaphor — locked services visible, upgrade modal on click.
Every freemium SaaS since 2012. Apple App Store in-app purchase (2009). Shareware paywalls (1985). Steam DLC. Not patentable.
Dependent Claims (4–18)
DEAD
Claim 4. 37-message boot-sequence animation with timestamps and status tags.
Purely decorative UI. Cinematic flair is copyrightable (as expression), not patentable (as method). See also: every indie web OS demo 2010–2024.
DEAD
Claim 5. Structured data cards per tool type (tabular, list, grid, metric).
Every dashboard framework since Material Design / Bootstrap cards. Stripe Dashboard, Grafana, Retool, Notion — all prior art.
WEAK
Claim 6. Dashboard widget grid (7 widget types) toggleable by NL command.
Widget grids aren't novel. The "toggleable by NL" hook is thin — Siri + HomeKit does this. Maybe survives as narrow combination claim, but trivial to design around by renaming widgets.
DEAD
Claim 7. Multi-round tool loop (up to 6 rounds), JSON result injection, terminate on non-
tool_use or all-errors.
This is literally the Anthropic Tool Use cookbook (2024). Public documentation = anticipating prior art. OpenAI function-calling loops since 2023. Specific "6 rounds" limit is obvious tuning.
DEAD
Claim 8. Credit metering: 500 / 5k / 50k / 999,999 per tier, JSON persistence, 1-credit per call.
API-credit metering is 25 years old (AWS, Twilio, every paid API). Specific numbers aren't patentable — they're just a price list.
DEAD
Claim 9. À-la-carte service purchase without full tier upgrade.
Literally the Apple App Store, Steam, every in-app purchase since 2008.
WEAK
Claim 10.
_action pattern — tools return client-directive objects (set_background, open_app, show_notification, etc.) instead of server-side execution.
Conceptually Redux / Elm Architecture (2013) — server returns "actions" for client to dispatch. BUT the specific combination of AI tool output → client-side OS control directive is narrower. Still probably obvious; doable but likely rejected on §103.
DEAD
Claim 11. Dual-model smart routing: keyword regex → heavy model (Claude) OR lightweight (Ollama).
Mixture-of-Experts + model cascades: published 2017+. LangChain RouterChain (2023). Every production LLM system does this. Not novel.
DEAD
Claim 12. Conversation persistence: 50-msg sliding window, JSON on disk, per-conversation ID.
Chat history with sliding window = every LLM chat app since 2023. File-backed JSON = obvious storage choice. §103 kill.
LIVE*
Claim 13. Tool-pill UI state machine: pending → animated → completed → checkmark, with write-op green-border / read-op yellow-border, keyed by execution ID.
The ONLY claim with a plausible narrow path through USPTO. Specific visual state machine tied to specific AI-tool semantics. Still arguably obvious (progress spinners exist) — but narrow enough to file as design patent. Cheap filing, limited scope, hard to enforce.
DEAD
Claim 14. Electron wrapper of the web UI with
electronAPI IPC bridge.
Every Electron app since 2014: Slack, Discord, VS Code, Figma, Notion, Atom. IPC bridges are in the Electron docs. Zero novelty.
DEAD
Claim 15. API proxy layer mapping
/api/<svc>/<ep> to 127.0.0.1:<port>/... with 5s timeout + 502.
Reverse proxy / API gateway: nginx (2004), Envoy, Kong, Traefik. Same-origin routing of microservices is textbook. §102 kill.
DEAD
Claim 16. Terminal emulator with AI-default mode,
$/! shell passthrough, built-in commands (help, apps, neofetch, etc.).
xterm.js + web terminals since 2014. Warp (2022) and Wave Terminal do AI-default mode. Replit and Coder.com ship similar. Built-in commands are trivial.
DEAD
Claim 17. File-watch live-reload dev mode via
fs.watch() + SSE /__reload.
webpack-dev-server (2014), Vite (2020), nodemon, LiveReload (2010), browser-sync. This is the default of every modern dev stack.
DEAD
Claim 18. Error classification into {timeout, rate_limit, auth, network, unknown} via regex on error message.
Every HTTP client library does this (axios, fetch wrappers, retry libraries). Regex taxonomy isn't a patentable algorithm.
Scoreboard
| Category | Count | % | Reason |
|---|---|---|---|
| ● DEAD | 15 | 83% | Alice §101 abstract-idea OR §102/103 prior art (2014–2025) |
| ● WEAK | 2 | 11% | Narrow combination claims, easy to design around |
| ● LIVE | 1 | 6% | Tool-pill UI state machine — possible design patent |
What Is Still Actually Patentable
⚠︎ Honest answer: very little of the original spec.
The patent draft was written in good faith in April when the market was less mature. Between April 8 and today (Apr 21), the world shipped: OpenAI Operator, Claude Agents, ChatGPT desktop apps with tool control, Anthropic Skills, Vercel v0 desktop mode, Replit Agent, Cursor Composer, Manus, Devin. Most of the "desktop OS + AI orchestrator" moat is now commodity.
Narrow slices that could still file (if you insist)
- Design patent on the tool-pill + data-card UI kit — $200–$500 filing, protects the specific visual appearance of the pills (shape, state transitions, border colors). Cheap, narrow, enforceable against literal copycats only. Does NOT cover functionality.
- Design patent on the boot-sequence animation sequence — same rationale. Protects the look, not the idea.
- Provisional utility patent on
_actiondispatch pattern — $320 pro-se. Establishes priority date for 12 months. Gives you "patent pending" marketing rights. Converts to a full utility in 12mo if you still care. Likely rejected at examination, but the provisional costs you one beer a month for 12 months.
What you should NOT bother filing
- Utility patent on the OS metaphor, AI orchestrator, or tier system — will fail §101 Alice, wasting $10k–$20k and 2–4 years
- Any claim mentioning "on a computer" or "using AI" as the novel element — Alice killed that phrasing category
- International filings (PCT) — $20k+, same §101 problems in EP/UK, stronger in JP/CN but not worth it at this revenue scale
The Real Moat (what to do instead)
Rank-ordered by $ / hour of effort:
- Trademark CASH.BOT — $350 USPTO, 20min online, protects the brand. File this week. tmsearch.uspto.gov first to check conflicts.
- Domain portfolio — already own cash.bot. Defensively register
cash.bot.com,cashbotos.com, common typos. <$100/year total. - Copyright — automatic on every file. For extra enforcement, register key works with US Copyright Office ($45 each) — usually unnecessary unless you intend to sue.
- Trade secret — the orchestrator prompts, customer DB, pricing funnel, revenue math. Keep in-house, NDA contractors, don't open-source. This is the strongest legal protection for software that Alice killed.
- Ship speed — 65 services in a year is a physical moat. Competitors need headcount you don't need to match.
- Distribution — cash.bot apex + home.cash.bot tools grid + community + WA list + email drip. None of this is patentable but all of it compounds.
- Integration bundle — OS + IDE + VPS + AI + community + casino + mail at $9 starter is a switching-cost moat. Harder to unbundle than to patent.
Action Items
Do
- File USPTO trademark for CASH.BOT this week ($350, uspto.gov/trademarks/apply)
- (Optional) File design patent on tool-pill UI ($500 total w/ filing + drawings)
- (Optional) File provisional utility on
_actiondispatch pattern ($320 pro-se) — only if "Patent Pending" in marketing is worth a year - Keep
apps/docs.cash.bot/PROVISIONAL-PATENT-FULL.mdas internal design doc — it's a decent technical spec even if not filed
Don't
- Hire a patent attorney for a full utility on any of the 18 claims — ROI is negative
- Mention "patented" or "patent pending" in copy until at least a provisional is filed — it's literal false advertising (35 USC §292)
- Spend money on international PCT filings at current revenue scale
References
- Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014) — killed most software business-method patents
- Bilski v. Kappos, 561 U.S. 593 (2010) — earlier swing at abstract-idea rejection
- 35 U.S.C. §101 — patentable subject matter (Alice analysis applied here)
- 35 U.S.C. §102 — novelty (prior-art anticipation)
- 35 U.S.C. §103 — non-obviousness
- 35 U.S.C. §292 — false marking ("patent pending" without filing = up to $500 / offense)
- USPTO Trademark Electronic Application System (TEAS): uspto.gov/trademarks