Where does my money sit?

On the exchange you connect. We never custody. The API key has no withdrawal permissions, and your exchange's permission UI proves it during onboarding.

Which exchanges do you support?

Kraken Futures is the primary integration today. Coinbase, Bybit, Binance, and OKX onboarding are in active rollout. The engine is exchange-agnostic.

What happens if a strategy stops working?

Each strategy's win rate is tracked over a rolling two-week window. If it drops, the system stops trading it automatically and flags it on your dashboard.

Can I add multiple exchange accounts?

Yes. Each account is fully isolated: its own keys, its own daily limit, its own kill-switch. They run in parallel; one going wrong never touches the others.

How fast does the kill-switch trip?

You set a daily loss limit. The moment the system hits it, no new trades open. Existing trades exit normally.

Does it learn from my trades?

Yes. Once you connect an account, the in-house model feeds your specific outcomes back into its scoring. Over a few hundred trades it tunes itself to the conditions you're actually trading in.

Crypto exchanges have the cleanest APIs for retail users and trade 24/7. Stock and futures brokers are coming next.

AlgoTrada is software that places trades for you. We're not a broker, custodian, or investment advisor. Your money sits at your exchange under their license.

Can I see how trades are decided?

Yes. Every trade comes with a one-paragraph explanation: which markets agreed, which strategy fired, what the historical look-alikes did. No black boxes.

designed roadmapgate-based · evidence over quarters

The honest read

One first market. Everything else is earned.

The litepaper describes the destination: an agent-first exchange with a marketplace and staged live execution. This document is the plan. A single wedge, gates sequenced by dependency and proof, each with a pass metric and a kill criterion. Calendar labels are targets; gates are the truth.

First market

The proving ground for trading agents

Not “an exchange” on day one. The range: agents rehearse against recorded markets at chosen speed and come back with receipts proving no look-ahead. Unique today, hard to copy honestly, and the exact thing live-venue agent APIs (Kraken, Binance, OKX, Robinhood) do not offer.

“TradingView bar-replay for AI agents: metered sessions, proof of no look-ahead.”

Buyer: the agent’s operator, meaning semi-technical traders running Claude/ChatGPT/Gemini, and agent-framework developers who need an environment and an eval. Distribution: an MCP connector + skills inside the chat clients they already use. Crypto tapes on the public tier; CME futures in private beta until licensing clears.

Operating principles

One product per gate.A seed team ships one thing at a time; every gate has a single headline deliverable.

Receipts are the brand.Every claim we make a user can verify; every feature that can't be receipted waits.

Agents don't pay; operators do.Pricing, onboarding, and docs address the human holding the API key.

Regulated things last.Marketplace, copy-trading, live routing sit behind both evidence gates and counsel.

Gates

Credible venue

in progresstarget · wks

The thing we demo must not wobble. Ship the fixes already in flight and lock them with the visual loop.

Ship

Layout fixes deployed (void, gate, full-height cockpit rail)
Live decision tape end-to-end (WS `decision` frames)
ASCII mobile containment; visual-verify green on deploy

Depends on

Backend `decision` WS variant (in flight)
Merged agent edits, one coherent deploy

PASS → visual-verify exits 0 on taifoon.io, full state matrix; a stranger completes one replay session unaided. Kill: nothing, this gate is table stakes.

Agent door + Cubicle skin

nexttarget · Q3 '26

Acquisition. One key, one MCP connector, docs an agent can follow, under the AlgoTrada brand.

Ship

MCP server + skills (sessions, orient, orders, receipts)
The Cubicle skin on the AlgoTrada site
Glyph-grammar wiki v1 · quickstarts for Claude/GPT/Gemini
Free tier: N session-hours/mo, crypto tapes

Depends on

G0 · domain + entitlement stub (free tier only)

PASS → 100 operators run ≥1 session; ≥25 come back in week 2 (retention, not signups). Kill/pivot: operators sample once and vanish → the wedge is wrong; re-aim at framework-eval (sell the environment to OSS agent frameworks) before building anything else.

First revenue: metered proving ground

target · Q4 '26

Charge for what is already valuable: parallel sessions, speed, longer windows, private tapes, receipt exports.

Ship

Billing + metering (session-hours × speed)
Skills packs v1 (indicators, graded answers)
Session forking / A-B compare · RL-friendly eval exports
Design-partner program (10 named operators/desks)

Depends on

G1 retention signal · measured $/session-hour
CME licensing path chosen (futures stay private beta until)

PASS → first $10k MRR or 3 design partners on paid contracts; unit margin on 1000× sessions known and positive. Kill: users love free, won't pay → move revenue to B2B eval licensing; do NOT proceed to marketplace on hope.

Identity: the keystone

target · with G2

Canonical subjects (author / agent / algo, legs resolved) replacing string-prefix attribution. Unglamorous; everything trust-shaped sits on it.

Ship

Canonical attribution resolver + subject API
Per-subject receipted track records
Leaderboard/account queries off prefix-matching

Depends on

Backend work (documented in acceptance docs)

PASS → zero cross-contamination in account queries (regression-tested); every session attributable to one subject. No kill, this is a prerequisite for G4/G5; if G2 failed, this work still serves B2B eval.

Marketplace & re-seller

target · Q1 '27 · earned

Only now: strategies listed with receipted track records, subscriptions, author payouts, copy-mirror with risk guards.

Ship

Listings + subscriptions (software/data framing)
Mirror executor: caps, kill-switch, audit chain
Author revenue share · Skydweller as house author №1

Depends on

G2 revenue signal · G3 identity · counsel per jurisdiction

PASS → 20 listed strategies with real subscribers; take-rate revenue > skills revenue trend. Kill: authors won't list / buyers won't trust → double down on eval + skills; the venue still compounds.

Staged live execution

target · Q2 '27 · earned

Paper → devnet → gated live through partner rails (Kraken, IBKR), armed-on-approval, receipts auditing every dispatch. The “exchange” story becomes literal only here.

PASS → first gated live accounts with zero unaudited dispatches. Kill: partner/regulatory friction too high → remain the proving ground + marketplace; that is already a company.

G∞

Open trading model

rides the corpus

Nemotron-family fine-tune on the receipted decision corpus. An option that accrues value at every gate, priced as research, never on the critical path.

What we deliberately do not do

No live capital before G5. No marketplace before identity. No futures-data marketing before licensing. No “AI-first exchange” headline before the proving ground retains. The bigger claim is earned at G4/G5, and saying it early invites comparison to venues with 1000× our liquidity.