AI Agent

The HyperionX chart AI is an assistant for the active chart. Use it to explain price and indicator state, inspect strategy context, troubleshoot the desktop, and prepare Code Lab work.

Feature status

Chart chat and its read-only context are Current. Structured desktop actions are Preview and simulation-first. Live Hyperliquid and KuCoin actions are available only in a release-enabled build that reports the matching provider capability and passes its local gates. The managed Rion cloud desktop bridge is Internal and is not a third-party integration API.

Rion Autopilot and Strategy Discovery are separate legacy Preview workflows. When that workflow is available and invoked, Strategy Discovery can upload the full research CSV and metadata by default without a dedicated consent prompt. See Rion Autopilot And Strategy Discovery.

Check The Conversation Engine

Reviewed 1.1.27 source

The behavior below is verified against reviewed desktop source, not a guarantee about an installed public release. Check Help > About, the available cloud capabilities, and the documentation version map.

In reviewed 1.1.27 source, HyperionX Cloud uses the grounded conversation engine by default. It can query chart history and coverage, inspect and change indicators, draw supported high/low levels from verified chart evidence, and create or revise a task-owned Code Lab indicator when permissions and entitlements allow. Check the returned tool result and the chart or compiler state before treating an action as complete.

This engine does not expose order-placement, Strategy Discovery, Validator, Optimizer, or Research Lab execution tools. A phrase such as find me a strategy does not launch the legacy discovery pipeline through this default Cloud path. Other configured providers and older builds can use different routing. A service that lacks the required conversation capabilities reports the limitation; it does not silently run the legacy workflow instead.

Choose A Provider

Open the AI settings and select the provider, model, and endpoint that should process the conversation.

Provider	Status	Current default	Configuration
HyperionX Cloud	Current and default	Rion	Uses the HyperionX license and cloud session; no customer model API key is required.
OpenAI	Current	gpt-4o-mini	Requires an OpenAI API key.
Claude / Anthropic	Current	claude-sonnet-4-6	Requires an Anthropic API key.
Ollama	Current	llama3.2	Uses http://localhost:11434 by default and does not require an API key. Tool support depends on the selected local model.
Azure OpenAI	Roadmap	—	Not enabled in the current provider registry.
Azure AI Foundry	Roadmap	—	Not enabled.
Foundry Local	Roadmap	—	Not enabled.
GitHub Copilot	Roadmap	—	Not enabled.
Copilot Studio	Roadmap	—	Not enabled.
Agent-to-Agent (A2A)	Roadmap	—	Not enabled.
Custom OpenAI-compatible	Roadmap	—	Not enabled.

Provider availability can also depend on the installed build, license entitlements, credentials, and model access. The chat header should identify the provider/model actually selected; verify it before sending sensitive market or strategy context.

Ollama is local only while its endpoint resolves to the local machine. If you configure a remote Ollama endpoint, prompts and context leave the desktop.

Check And Top Up Rion Credits

Rion credits apply to the managed HyperionX Cloud provider. They are not a broker balance, trading collateral, cash, or the billing meter for a customer-owned OpenAI, Anthropic, or Ollama connection.

To check the current cloud usage summary:

1. Open the chart AI panel.
2. Open the three-dot Rion menu.
3. Read the used and left credit status.
4. Select Refresh credits to request a fresh summary from the HyperionX gateway.

A valid Rion license and active HyperionX cloud session are required. Credits temporarily unavailable means the desktop could not retrieve the current summary; it does not establish a zero balance or prove that a new request will be accepted.

Select Top up credits to open the billing URL supplied by the gateway. This opens an external browser page; it does not purchase credits automatically. Before signing in or paying, verify the destination, account, amount, and current terms. After a completed top-up, return to the Rion menu and select Refresh credits.

Credit availability is only one cloud-access gate. The current license, plan, service status, and request limits can still affect whether a Rion request runs. For a customer-provided AI provider, review usage and billing with that provider instead.

Understand The Active-Chart Context

The assistant can build a snapshot containing:

• symbol and timeframe
• total, recent, and visible candles
• loaded indicator names, plots, signals, and selected parameters
• loaded strategy names, state, account label, positions, and selected parameters
• Chart Trader visibility and connection state
• displayed order type, size, prices, leverage, position, entry, and PnL text
• Validator, Optimizer, or Research Lab state when requested and allowed
• local chart-memory summaries

The snapshot follows the active chart. It does not currently target an arbitrary chart by public chart ID.

Keep Secrets Out Of AI Context

The current parameter filter omits names containing key, secret, password, or token. This is a defensive filter, not a guarantee that arbitrary custom text is safe.

Never place these values in a prompt, script parameter, profile instruction, prompt file, or feedback note:

• API or broker credentials
• passwords or bearer tokens
• wallet private keys
• seed or mnemonic phrases
• private customer or account information that the selected model provider should not process

HyperionX does not intentionally read broker credential fields into chart context. User-created parameter names and values can still contain sensitive information. See AI Data and Privacy for provider destinations, local paths, and retention.

Ask Read-Only Questions First

Useful requests include:

Explain what changed over the visible chart range and cite the candle or indicator evidence.

List the loaded indicators and explain what each contributes.

Explain the selected strategy state without placing or modifying an order.

Review this Code Lab compiler diagnostic and suggest a HyperionX-native correction.

Treat model output as a draft. It can misread context, invent APIs, or produce unsafe trading logic.

Ask About An Image

In reviewed 1.1.27 source, paste an image into the chat input, check its preview, and add the question you want answered. Remove an unwanted attachment before sending. Image requests require the updated, vision-capable HyperionX Cloud service; the other provider paths do not accept them in this implementation.

Images are saved with protected local chat history and can be included again in follow-up cloud requests within the conversation's image limits. Crop or remove private account details and unrelated content before pasting. Text secret filters do not redact pixels. See AI Data and Privacy.

Use AI With Code Lab

In reviewed 1.1.27 source, the default Cloud conversation can read the HyperionX authoring API, create a task-owned indicator, revise its source, compile it, inspect actual diagnostics, and open the artifact in Code Lab. This requires the relevant profile permissions and Code Lab entitlement. A build request can include bounded compiler-repair attempts; inspect the final compiler result rather than relying on the model's description.

Generated indicators use Indicators\Rion\<ClassName>.cs beneath the configured custom-project root. Rion rejects an existing unowned class filename and detects external edits before revising a file it created. This is scoped artifact authoring, not arbitrary access to edit every script. Compilation does not establish correct plotted behavior or successful runtime validation. Review the source and test its outputs on the intended data and session definitions.

The separate Local Desktop API route POST /api/v1/rion/code-lab/action supports opening Code Lab, listing scripts, compiling the custom project, and reporting compile state. That scoped route does not create or edit source and does not validate one arbitrary file. Providers or builds without the grounded artifact tools can still draft source for a manual Code Lab handoff.

Legacy Rion Autopilot and Strategy Discovery have their own generated-file, research, Validator, and Optimizer workflows. They are not the default Cloud conversation's indicator-authoring tools. Review their routing, trigger phrases, opt-outs, and consent limitation before using an installed path that supports them. Use local only or no ml when running legacy discovery with research data that must remain on the workstation, and verify the ML stage was skipped.

For generated code:

• use the documented HyperionX.Custom namespace and base for the extension type
• use OnStateChanged() and OnBarUpdate() where required
• expose settings with [HyperionXProperty]
• guard early bars and every added series
• prefer managed strategy-order helpers for normal order flow
• review direct SubmitOrder(...) use for cancellation, OCO, partial fills, and state transitions
• test in historical validation, playback, and Hyperion Sim before considering a live workflow

Start with the Developer Center for supported extension contracts.

Trading Actions

The default grounded Cloud conversation in reviewed 1.1.27 source has no trading-execution tool. An order request in that chat must not be treated as an order submission or as entry into the legacy confirmation flow.

Separate legacy in-app and Local Desktop API paths can support permissioned Hyperion Sim actions for placing, canceling, modifying, closing, or flattening orders. Their availability must be checked in the installed build; see the documentation version map.

Newer release-enabled builds can expose guarded Hyperliquid and KuCoin live actions. A caller-provided confirmed or executeTrading field records the caller's approval decision; it does not create an independent desktop confirmation prompt for a direct HTTP request. The client must also pass the provider-specific capability, setting, entitlement, active account/instrument, connection, and exchange-permission checks.

Use this progression:

1. Inspect read-only context.
2. Draft and review code or an order idea.
3. Compile and validate scripts.
4. Run in historical testing or playback.
5. Use Hyperion Sim with an explicit account and user decision.
6. Check the returned action status and authoritative account state.
7. Consider live Hyperliquid or KuCoin only in a release-enabled build and follow the Local Agent Live Trading API checklist.

See Local Desktop API for the Preview route and permission boundary.

Use A Specialized Profile

Agent Builder creates specialized profiles with selected context and tools. Some profile toggles are enforced for non-baseline specialized agents; many displayed trading-risk fields are currently advisory metadata. Agent Builder profiles also do not scope the global Local Desktop API token.

Review Agent Builder API before treating a profile setting as a security boundary.

Clear Conversation History

In reviewed 1.1.27 source, right-click inside the chart chat panel and select Delete all to start a fresh conversation. This removes messages and attachments from the saved chat history, clears pending input and confirmations, resets the conversation's local tool context, and restores the welcome message.

The desktop also attempts to clear the previous cloud conversation context when a valid session is available. That cleanup is not a guarantee of deletion from the model provider's retention systems. Generated Code Lab files and separate database chart memory are not deleted by this command; chart memory has no dedicated purge control in the AI UI. Older builds may use the label Clear Chat.