There is no hosted public instance and no signup. You run it. That takes about five minutes and needs no credential.

​
1. Clone and start the dashboard

1

Clone

2

Install the dashboard

--ignore-scripts is required on Windows, where native postinstall steps fail. A zero exit code from npm install does not by itself mean every dependency built.
3

Run it

The dashboard binds port 3101. Run it from apps/dashboard — there is no root workspace, and starting it from the repository root makes every route 500 through a Tailwind config-resolution failure that looks exactly like a broken change.

​
2. Your first call

Expected shape
Note read_this_first: the response states its own epistemic status in prose, in the payload. That is deliberate — a payload travels to places documentation does not.
data_mode: "empty" is the correct first response, and it does not mean “no risks found.” It means no agent run has happened on your machine yet. This distinction is the single most important thing to internalise before building against this API — see Data states.

​
3. A call that returns real data immediately

The provider leaderboard ships with a committed board, so it returns content on a fresh checkout:
This is the middle of the board’s three data modes, and the reason there are three rather than two. The board is measured — nothing here is demonstration data — but total_scores is 0, so there is nothing to read yet. Collapsing that into “measured” alone would be technically true and actively misleading, so the payload distinguishes it in prose. A response that states its own epistemic status is the pattern to expect everywhere in this API.

​
4. Verify something

Every signal can be independently re-checked:
This re-derives the evidence hash from the published payload and confirms the model cited only what it was shown. It cannot prove the evidence was true when collected — see Verify a signal for exactly where that line sits.

​
5. Pick your next step

Try It in the browser

Every public endpoint has an interactive playground pointed at your local instance.

Connect an AI agent

38 read-only MCP tools in Claude Code, Claude Desktop, Cursor, or Windsurf.

Score a real trader

Three minutes, using live Hyperliquid data and no credential.

Understand the numbers

How each figure is produced, and what it cannot show.

​
Optional: the self-hosted engine

The Python engine is a separate surface with its own endpoints (/edge.json, /risk.json, wallet scorecards, backtests, execution):
It binds 127.0.0.1:8777 and requires no dependencies beyond the standard library. See Self-hosted endpoints.
If you expose it through a tunnel, set tunnel_token in state/config.json first. With no token configured it is open to anyone who can reach the port — safe on loopback, unsafe the instant a tunnel is running. Nothing prompts you. See Authentication.
Scienter publishes impersonal market signals of general and regular circulation. Nothing here is investment, legal, or tax advice, or a recommendation to buy, sell, or hold any asset. Scienter is not a registered investment adviser or broker-dealer. Trading digital assets can result in the total loss of your funds — see Disclaimers for the full text.