Dashboard
kalshi-ai v1 · design · spec updated 2026-09-20
Kalshi
Live from the exchange API. Most active open markets right now.
Agent pipeline
Design state from the spec. Runs are background jobs with hard caps; nothing here is live yet.
Find markets worth attention across all open markets
- model
- cost-efficient
- runs
- scheduled; delayed for free tier, fresh for paid tiers
- output
- ranked shortlist with one-line rationale per market; shared across all users
Form an independent probability for one market
- model
- frontier
- runs
- on demand, or automatically for paid watchlists
- output
- probability, confidence, edge, evidence with citations, rules reading
Turn an analyst result into a recommendation
- runs
- after each analyst result
- output
- position, size_guidance, entry_price, exit_conditions, invalidation, explanation
Milestones
- M1vertical sliceup next
Kalshi adapter, one market analyzed end to end, recommendation logged with tracked outcome
Excludes: auth, billing, UI polish
validates the interesting part before any infrastructure that depends on it
- M2scanner and track record
- M3auth, tiers, credit ledger
- M4stablecoin payments
- M5polymarket adapter and cross-venue signals
Decisions
Version 1 is analysis and recommendation only. No execution.
Users act on recommendations themselves. The app deep-links to the market on the venue. The provider interface excludes execution methods entirely so the boundary is structural, not a config flag.
Stablecoin recurrence model
- prepaid_periods
- credit_top_ups
- both
Card fallback ever, later, or never
Legal counsel engaged, or draft the questions to bring to one
Free tier split between shared computation and per-user quota
Lean: shared scanner output plus a small personal quota
Pricing for standard and elite
Risk register
| ID | Risk | Mitigation | Status |
|---|---|---|---|
| R1 | Personalized paid recommendations may look like Commodity Trading Advisor activity | Fintech counsel before taking payment; research-and-analytics framing; disclaimers | open |
| R2 | LLM misreads resolution criteria | verbatim rules in output; explicit edge-case reading; citations required | mitigating |
| R3 | Unbounded agent cost on free users | hard caps per run; credit check before job start; shared scanner output | mitigating |
| R4 | Stablecoin-only checkout limits conversion | decide on card fallback timing; see D3 | mitigating |
| R5 | Venue API terms may restrict advisory use | review Kalshi and Polymarket developer terms during adapter work | mitigating |
Tiers
- · browse and search all markets
- · delayed scanner shortlist
- · small monthly credit grant for deep analyses
- · public track record
- · daily fresh scanner
- · meaningful monthly credit grant
- · watchlists
- · resolution alerts
- · private history
- · real-time scanner
- · cross-venue arbitrage feed
- · auto-analysis on watchlist changes
- · API access
- · larger credit grant
Stack
- web
- Next.js 16.3.5
- language
- TypeScript
- db
- DynamoDB
- jobs
- Inngest or Trigger.dev
- llm
- Claude API with tool use
- auth
- clerk
- payments
- stablecoins only
- apps/web
- packages/providers
- packages/agents
- packages/db