Skip to main content

AI Data and Privacy

Use this page before enabling Rion, connecting a third-party model provider, sharing an AI-related support bundle, or giving an Agent Builder profile access to chart and script context.

Know the destination

HyperionX Cloud, OpenAI, and Claude are cloud destinations. Ollama is local only when its configured endpoint is local. Selecting a provider determines where prompts and supplied context are processed.

Strategy Discovery can upload research data without a dedicated consent prompt

The legacy Rion Strategy Discovery workflow enables its cloud ML stage by default. When that workflow is available and invoked, it can upload the full research CSV and associated metadata to the HyperionX-managed ML service without first showing a dedicated consent prompt. This is a blocker for unrestricted public-release use. Add local only or no ml to the discovery request whenever the dataset must remain on the workstation, and verify that the ML stage was skipped. The default grounded Cloud conversation in reviewed 1.1.27 source does not launch this workflow. Read Rion Autopilot And Strategy Discovery before using an installed legacy path.

Image and conversation-reset behavior labelled reviewed 1.1.27 source below is not a claim that every installed release includes it. Check the documentation version map and the installed build and cloud capabilities.

What Can Be Added To A Prompt​

The chart assistant builds context from the active HyperionX workspace. Depending on the request and the selected profile, that context can include:

  • symbol, timeframe, and candle data
  • indicator names, plot values, signals, and selected parameters
  • strategy names, state, account labels, positions, and selected parameters
  • Chart Trader state, including displayed size, leverage, entry, and PnL text
  • open-window and research-tool state
  • recent local chat history and chart-memory summaries
  • Code Lab script names, paths, source context, or compile diagnostics when the relevant tool is allowed
  • pasted image attachments and retained images from earlier turns in the reviewed 1.1.27 Cloud conversation path

This information can be commercially or personally sensitive even when it contains no credential. Review account labels, script parameters, strategy names, and prompts before sending them to a cloud model.

Provider Data Flow​

ProviderStatusProcessing destination
HyperionX Cloud / RionCurrentHyperionX-managed cloud service. Uses the signed-in Rion license session.
OpenAICurrentOpenAI service configured by the user.
Claude / AnthropicCurrentAnthropic service configured by the user.
OllamaCurrentThe configured Ollama endpoint. The default is http://localhost:11434; a non-loopback endpoint is not local processing.

HyperionX controls the context it prepares, but the selected model provider controls service-side processing and retention. Consult the provider terms that apply to the account and endpoint you configured. HyperionX documentation does not override those terms.

Images And Follow-Up Requests​

In reviewed 1.1.27 source, pasted images require the updated, vision-capable HyperionX Cloud conversation service. Inspect the attachment preview and remove unwanted images before sending. Other configured provider paths reject image attachments in this implementation.

Sent image data is retained in the protected local chat-history payload within storage limits. Earlier images can be sent again on later turns within the conversation's image limits, even when the follow-up message contains only text. Starting a fresh conversation removes that chat's retained attachments from its saved history; it does not revoke images already processed by a cloud provider.

Crop or obscure private account information, credentials, customer data, and unrelated windows before attaching an image. Text redaction does not sanitize image pixels.

Secret Filtering Is A Safety Layer, Not A Vault​

Before script parameters are added to chart context, the current build omits names containing key, secret, password, or token. Other feedback and memory paths apply additional pattern-based redaction.

Do not rely on those filters to classify arbitrary custom data. In particular:

  • do not put seed phrases, mnemonic phrases, private keys, bearer tokens, or passwords in script parameters
  • do not paste credentials into prompts, profile instructions, prompt files, or feedback notes
  • use neutral account labels that do not reveal information you would not send to the selected provider
  • review generated scripts before compiling or sharing them

The chart context is designed not to read broker credentials or wallet-key fields. A custom parameter or prompt can still contain sensitive text supplied by the user.

Data Stored On The Desktop​

Paths below use standard Windows environment variables.

DataLocationProtection and retention
AI provider settings%APPDATA%\HyperionX\AIAgent\settings.jsonProvider, model, and endpoint settings are stored locally. API-key content is protected with Windows DPAPI for the current Windows user.
Rion license session%APPDATA%\HyperionX\AIAgent\license.jsonAccess, refresh, and license-session secrets are protected with Windows DPAPI for the current Windows user.
Chat history%APPDATA%\HyperionX\AIAgent\chat-history.jsonUp to 80 recent messages are stored in a DPAPI-protected payload. Reviewed 1.1.27 source also retains image attachments within storage limits. Right-click Delete all removes the conversation history and attachments and writes the welcome state; older builds may label this Clear Chat.
Agent Builder profiles%APPDATA%\HyperionX\AIAgent\agents.jsonProfile instructions, permissions, workspace paths, and other non-key configuration are stored as readable JSON. Provider API keys are global settings and are not included in profile exports.
Chart memory and AI feedback%USERPROFILE%\Documents\HyperionX\DB\HyperionX.db3Stored in the main application SQLite database. Chart memory keeps up to 80 snapshots per symbol/timeframe, with up to 240 candles in each saved snapshot. The current UI has no dedicated chart-memory purge control.
Local API audit%APPDATA%\HyperionX\AIAgent\Audit\rion-api-yyyyMMdd.jsonlDaily JSON Lines files. Request bodies are replaced with a redacted length marker; metadata, result status, and messages remain readable. No automatic public retention schedule is currently documented.

DPAPI protection is tied to the current Windows user and machine context. It is not a portable backup format, and it does not protect data from software already running with that user's authority.

Managed Rion Desktop Bridge​

The managed Rion desktop bridge is an Internal HyperionX component, not a public integration API. With a valid HyperionX Cloud session, it can receive an approved managed command, invoke a permitted local Rion action, and return the action result to HyperionX Cloud.

Do not build third-party integrations against the bridge or depend on its transport. Use the Preview Local Desktop API only when HyperionX has provisioned the integration.

Review And Remove Data​

  1. In reviewed 1.1.27 source, right-click the chart chat panel and select Delete all to remove its saved messages and attachments. Older builds may use Clear Chat.
  2. Review Agent Builder prompt files, skill folders, and export archives before sharing them.
  3. Treat the HyperionX application database and AI audit directory as sensitive user data.
  4. Review support exports before sending them. They can contain symbols, account labels, local paths, error messages, and strategy information even when request bodies are redacted.
  5. Use the model provider's own controls for provider-side history or retention.

Delete all also clears pending input, pending confirmations, and local conversation tool context, and starts a new conversation. The desktop attempts cleanup of the previous cloud conversation context when a valid session is available; this is not proof that provider-retained data was erased. It does not delete generated Code Lab artifacts, separate chart memory, audit files, research datasets, or reports.

For generated-code, research-upload, report, and watchlist boundaries, see Rion Autopilot And Strategy Discovery. For local automation boundaries, see Local Desktop API. For profile-level context controls, see Agent Builder.