Loading…
Pre-seed · 2026
An autonomous trading operator, built on Hyperliquid.
Hyperion runs a trading desk around the clock. It ingests markets continuously, reasons about them in writing, and executes toward long-term financial goals you state in plain language.
On-chain markets solved access. Anyone, anywhere, can hold a position on a venue open around the clock, with custody in their own hands.
Attention is what's still missing. Someone still has to read the order book, funding, news, and flow at three in the morning, and still has to be at the screen when a limit order needs replacing before the market moves past it.
The value an agent actually adds is capacity: the ability to ingest order book state, news, and liquidity flows continuously, and hold all of it against a mandate, a goal, a horizon, and risk limits a trader already believes in.
It's the medium a thesis gets expressed through. You say what you want to hold, over what period, at what risk, and the agent handles the watching, the charts, and the limit orders.
State the outcome, delegate the mechanics. The onboarding surface of trading is about to become language, because the execution surface can run on its own now.
This pattern is spreading beyond trading. Travel platforms are shipping agents that search, book, and pay for a trip in one pass, with the transaction handled end to end by software. A crypto wallet lets software hold funds and move them directly, bypassing KYC at a bank, so agents already show up as counterparties on-chain.
Industry estimates already put the agentic economy in the trillions of dollars of annual activity by the end of the decade. Markets are the sharpest edge of it: liquid, quantifiable, and open around the clock.
We think trading is the leading edge of the agentic economy. Hyperion is built for that edge first.
Hyperion is that operator: an autonomous trading desk sitting between a trader's mandate and the venue. We're building on Hyperliquid because it's the first on-chain venue with the performance and liquidity to make continuous, serious execution possible.
Trading is the sharpest edge of a much larger shift. The same mandate-and-delegate pattern is starting to run commerce, procurement, and services. Here's how we size it, from broadest to what we can actually capture first.
The broad shift: agents executing on stated intent across commerce, finance, and services by 2030. Midpoint of PwC ($2.6–4.4T annual GDP contribution from agentic AI) and McKinsey (~$13T in additional output, $3–5T of that in agentic commerce alone).
On-chain spot and derivatives volume, extrapolated from current trajectory. Hyperliquid alone has cleared $4.4T in cumulative perp volume to date and runs at a multi-hundred-billion-dollar monthly clip (DefiLlama); on-chain venues broadly scale with it as agentic participation grows.
Subscription + bps on flow from Hyperliquid's prosumer trader base stating mandates in the first 12–18 months. A sliver of SAM, sized to what one execution layer can actually capture early.
Order-of-magnitude estimates we derived ourselves and applied specifically to Hyperion, built from published agentic-AI economic-impact research (PwC, McKinsey) and on-chain volume data (DefiLlama).
Order books, funding, open interest, positions, and flow on Hyperliquid, normalized into one live picture of the market, around the clock.
The picture is weighed against your mandate (horizon, targets, risk limits). Every decision is reasoned in writing before it's acted on.
Orders are sized, staged, and placed on Hyperliquid through hard-coded risk gates. Results feed back into the picture, and the loop continues.
The loop is inspectable end to end. You can read every decision the agent has made, and you can stop it at any time.
ILLUSTRATIVE CONCEPT · PRODUCT IN DEVELOPMENT
The full loop (ingest, reason, execute, journal) exists and runs today. It de-risks the hard parts (signing, gating, continuous reasoning) and generates the journal evidence the product story rests on.
Live aggregation across 10–30 Hyperliquid markets: multi-timeframe bars with perp-native metrics, including CVD, basis, funding trajectory, OI delta, liquidation proximity, and cross-asset correlation.
Model-agnostic (Anthropic / OpenAI / DeepSeek). Timeframe-batched digests in; schema-validated trade candidates with written theses out. Structured output, always.
Every order passes compiled risk gates: max position, max exposure, max concurrency, price sanity against live mark, post-stop cooldown, and a daily-loss kill-switch.
The master key signs exactly one approveAgent transaction. The daemon holds a scoped agent wallet restricted to trading. Signing logic owned in-house.
Append-only. Every candidate, thesis, and fill in one place: audit trail, backtest corpus, and the agent's memory.
Any agent that speaks MCP can read markets and place orders through the same gates. Every client shares one path to the wire.
Hyperliquid became the dominant on-chain perp venue, moving billions a day through a public, signature-gated API you connect to directly.
Every major lab shipped tool-calling agents, and MCP standardized the way they plug into the world, all within the same 18 months.
What's missing is trustworthy execution and a mandate-level interface. The models are ready; the wire to the venue is still being built.
The curves cross exactly at agents that trade.
Verifiable, append-only decision records: the reputation layer autonomous trading will need.
Web app, MCP clients, and the autonomous loop share one executor and one set of compiled gates. Auditable by construction.
The dangerous layer is ~300 lines of code, written and owned in-house, tested byte-exact against reference vectors.
Competitors ship bots (static strategies) or copilots (chat over charts). Everything here is built around the mandate: goal, horizon, risk envelope, written judgment.
Ingest → reason → execute → journal running end to end, with the MCP interface shipped.
Mandates in plain language, a readable decision log, one-click scoped-wallet onboarding, halt at any time.
Scoped signing as a service, mandate reputation built on verifiable journals, multi-venue routing.
Copied to clipboard
Structured rendering of this page's content, for models and other machine readers.
Loading…