Skip to main content

Agents

An agent is an isolated configuration with one assigned trading account, separate analysis and trading instrument sets, sub-agent prompts, and scheduled sub-agents.

Configure and operate an agent​

Scheduled trading requires an active enabled agent with an assigned account, at least one selected perpetual instrument, at least one enabled Analysis job, an enabled Trading job, and an explicit model for every enabled scheduled job. HyperVibes otherwise reports Paused or Setup required.

The agent navigation has Analysis, Trading, and Review pages beneath a non-clickable Sub-agents heading. Analysis is a multi-instance management page; Trading and Review are singleton configuration pages. There is no separate market-analysis role or standalone Prompts page.

Analysis uses approved data tools and publishes scoped research memories. Trading synthesizes the current research context, records trading_decision audit memories when possible, and manages orders. Review records outcomes and learnings and can revise Trading and Analysis prompts when granted that capability. Missing or stale research does not block the scheduler or order gateway.

The interface also shows positions, orders, transactions, memories, conversations, and Indicators. Missing or stale live account data is marked unavailable; new agent-originated exposure fails closed until required live streams are current.

Reset memories​

The agent settings page provides a synchronous Reset memories action that deletes all memories for the agent after confirmation. It does not change sub-agent prompts or account settings.

Instrument selection​

Analysis instruments limit the markets Analysis and indicators can inspect. Trading instruments are the strict allowlist for new order exposure. Adding an Analysis instrument never automatically adds it to the Trading allowlist. An Analysis run explicitly granted the Trading-instrument capability may promote an active Analysis instrument or remove a Trading instrument. Removing the final instrument pauses new exposure without closing positions. At least one Trading instrument must be selected before the agent can trade.

Indicators​

The Indicators tab manages agent-owned, versioned PineScript-subset programs. Definitions target an explicit subset of the current analysis instruments and run only on server-fetched closed Hyperliquid candles. The tab shows recent status and values, supports PineScript text import, and charts the latest successful run. Editing creates a new immutable version rather than changing historical source or output. Loop statements are not part of the supported subset because the embedded Pine runtime cannot enforce an instruction budget.

The latest successful chart renders active plotshape(), plotchar(), and plotarrow() calls as markers derived from those same server-fetched closed candles. Shape and character markers support Pine text, colors, sizes, above/below/top/bottom/absolute locations, and positive or negative offsets that remain within the stored candle history. Lightweight Charts approximates Pine styles with circle, square, and up/down arrow markers; text color cannot be separate from marker color, and plotarrow() uses a fixed size rather than a value-dependent pixel height. All markers attach to the candlestick series, including indicators that are not overlays. Indicator markers are visual outputs only and never place orders.

Chat loads the pine-indicators authoring skill before creating or modifying an indicator. Review can use the same skill when its run has indicator-write capability and evidence supports an immutable revision. Analysis does not edit indicators; it consumes their numeric plots and marker events as research evidence and publishes its conclusions as memories for Trading.