Roadmap

What we're building next. Grouped by time horizon. This is a snapshot, not a commitment — priorities shift as we learn what matters.

Current date: 2026-04-20

Next 30 days — shipping soon

security Bearer-token auth on /api/*

Add CRON_SECRET env var. Enforce Authorization: Bearer <secret> on all /api/* endpoints. Vercel cron scheduler automatically includes the header; manual callers get the secret out-of-band. Closes the "anyone who knows the URL can trigger scans" gap called out in the security notes.

Real-exchange adapter (paper-trading first)

The api/exchanges/* adapter layer exists; wire it to a real paper-trading testnet (Binance Futures testnet or Bybit demo) so we can validate that the risk-sizing chain produces sane order sizes under real latency + partial fills.

Grafana dashboard

Per-cron success rate, latency histograms, model-call timing, Redis write volume. Targets the self-hosted ops use case.

Next quarter

Prop fund mode

A mode where Velox manages a single-user discretionary account: the operator confirms every consensus signal before it becomes a real trade. Bridges the gap between "watch the pipeline" and "trust the pipeline."

Alternative-data ingestion

Discord sentiment: already wired in but needs broader channel coverage and spam filtering
On-chain flows: large-holder movements from Glassnode / similar
Funding-rate history: currently spot-only; adding historical basis for context

Richer optimizer

Multi-objective optimization (Sharpe + Calmar + max drawdown)
Regime-conditional weights (weights that vary by detected regime)
Per-symbol weights (some models are better at majors vs alts)

Next 6 months

Fourth model (experimental)

Adding a fourth model changes consensus tiering:

2/4 → weak signal, log only
3/4 → silver
4/4 → gold

The question we want to answer: does adding an extra model improve precision (fewer false positives) or just slow the pipeline? Will A/B test for 60 days.

Cross-asset signals

Right now: 80 crypto perps. Extending to:

US equity index futures (ES, NQ) — same consensus engine, different universe
Spot gold and silver
Major FX pairs

All three models can already reason about these; the engineering is mostly data-plumbing.

Public API keys

Paid tier for programmatic access. Hits the /api/* endpoints with per-caller rate-limits and billing. Depends on the bearer-auth work landing first.

Explicit non-goals

We're deliberately not building:

A mobile app. The terminal is a desk tool; shipping native mobile is 6 months of work for marginal return.
A signals Telegram/Discord bot. Easy to build; easy to abuse (pump-and-dump groups already do this poorly, we don't need to add to that).
A copy-trading product. The legal surface area (especially cross-jurisdiction) is too wide for a small team.
Support for every exchange. Adapter layer supports plugging new ones in, but the maintained set stays small (Binance, Bybit for now).

Known debt

Stuff we're aware of and have deferred:

openai is actually Gemini in Redis keys, variable names, DOM IDs. Fixing requires a Redis migration script; cost exceeds benefit for now.
Cooldown TTL mismatch: scan.js sets 48h, cooldown.js sets 24h on the same key. First writer wins.
Optimizer disconnect: optimize.js tracks cooldownHours but scan.js hardcodes 4h. The optimizer value isn't consumed.
No conflict resolution between browser and server writes. Last-write-wins in Redis. Noisy in practice but hasn't caused a lost trade yet.

How to influence this list

Open a GitHub issue at chasing-labs/veloxtrade with the proposal label. The most valuable proposals are the ones with concrete user stories ("I want to do X so that Y") rather than technical prescriptions.