agent-first trading platform

Strategies your agent can actually trade on. 

Tested strategies, evidence-gated signals, and a fail-closed broker gateway, wired to Claude, Hermes, OpenClaw, or any MCP agent through one hosted URL. Nothing trades without your seal.

one live endpoint. 25 MCP tools for your agent, the same REST API for you.

walk-forward, out-of-sample ·25 MCP tools + REST API ·look-ahead structurally impossible ·plain English to a tested rule ·fail-closed broker gateway ·Ed25519-signed approvals ·21 verified marketplace strategies ·the platform never places a trade ·

why nakagai

Evidence first. Everything else second.

Evidence-gated signals

Every intraday signal ships with its own track record: out-of-sample windows, win rate, profit factor. Setups without validated evidence get suppressed, never dressed up.

A tester, not a promise

Walk-forward backtests with look-ahead made structurally impossible. Parametrize any of 13 strategy templates inside guardrail bounds, A/B two param sets over identical windows, and keep only what survives.

Strategy Foundry

Hand the Foundry a template and it searches the whole param space for the best win rate on a holdout-protected window, then certifies the winner on months it never touched before you run it live. Or just describe a setup in plain English and it compiles a tested rule.

A marketplace with receipts

Browse 21 seeded strategies, copy the ones you like into your workspace, comment, and trust the green check: every listing is re-tested on one standardized window, so verified means verified.

Fail-closed broker gateway

Your agent reaches Robinhood, Alpaca, or Coinbase through Nakagai, not around it. Connectors are off by default, only allowlisted tools are reachable, and anything not proven read-only counts as a write and gets refused.

Your seal on every trade

Nothing executes without your stamp. Approvals are Ed25519-signed, the agent literally cannot approve its own order, and you set a mandate for how far it can go and kill the whole thing in one tap.

how it works

Your agent does the work. You hold the seal.

step 1

Connect your agent

Point any MCP agent or plain HTTP client at one hosted URL. Claude, Hermes, OpenClaw, or your own code gets signals, strategies, backtests, the marketplace, and the broker gateway the moment it connects.

step 2

It scans, tests, and proposes

The scanner walks your watchlist, the Foundry tunes and certifies, and your agent sees only evidence: out-of-sample windows, win rate, profit factor. It brings you a proposal, not a pitch.

step 3

You hold the seal

Trades move only when you stamp them. Set a mandate for how far your agent can go, watch every check-in in the activity feed, and kill it all in one tap. Fail-closed, human-approved, agent-executed.

0

strategy templates

0

agent tools on one URL

0

verified strategies

0

trades the platform places itself

pricing

Free to watch. Pro to run the desk.

Free

$0/mo

Everything you need to watch the tape with an agent at your side.

  • Signals, watchlist, and scan health
  • Backtest history and run playouts
  • Marketplace browsing and copying
  • Agent access over MCP (read tools)
  • Strategy Lab templates + import/export

Pro

$49/mo

Waitlist members get their first 7 days of Pro free.

The full desk: test anything, optimize, and wire in your connectors.

  • Everything in Free
  • Launch backtests with custom params
  • Strategy Foundry: optimizer plus holdout certification
  • Connector management (brokers, data, notify)
  • Agent Mandate: woken-not-waiting autonomy
  • Communication channels (iMessage, notify connectors)

Full comparison on the pricing page.

Give your agent a seat on the exchange.

One hosted endpoint. Evidence-gated signals, a fail-closed gateway, and your seal on every trade.

Waitlist members get their first 7 days of Pro free.