Agent Builder

Agent Builder is the HyperionX environment for creating specialized AI agents inside the trading platform.

The goal is to move beyond a single generic chat assistant. Traders should be able to create agents for specific jobs, give them safe context, connect them to approved tools, and control what they are allowed to do.

Agent Builder is available from the main window and from Rion chat where the installed license includes the agent_builder entitlement. If the entitlement is missing, HyperionX should show a locked message instead of opening the builder.

Current Status

Current surface:

• Main window entry point.
• Rion chat entry point.
• License gate for Pro, Elite, or Admin users with the Agent Builder entitlement.
• Workspace restore support for the Agent Builder window.
• Agent configuration direction for model, tools, memory, prompts, permissions, and trading guardrails.

Production hardening still needs to keep the UI, tool permissions, prompt files, audit review, and trading settings aligned with the current Rion local API.

Agent Types

Examples of useful HyperionX agents:

Agent	Purpose
Code Lab Agent	Drafts, explains, debugs, and validates HyperionX scripts.
Strategy Research Agent	Reviews backtests, optimization results, and parameter behavior.
Chart Analyst Agent	Explains chart state, indicators, visible candles, and trade setups.
Risk Agent	Watches account state, open positions, order exposure, and configured limits.
Journal Agent	Converts trades and notes into structured journal entries.
Support Agent	Helps users understand platform windows, settings, errors, and logs.
Client Account Agent	Monitors multi-account state with restricted read access.

Agent Configuration

An agent should have a clear configuration model:

• Name.
• Description.
• Role.
• System instructions.
• Allowed tools.
• Allowed data sources.
• Workspace root.
• Memory policy.
• Model/provider selection.
• Risk limits.
• Trading permissions.
• Audit logging.
• Enabled/disabled state.

Workspace Root

Agents need a controlled workspace root for files they are allowed to read or write.

Recommended default:

~/.hyperionx/agents/workspace

The exact path should be configurable in platform settings. Agents should not receive unrestricted filesystem access by default.

Prompt Files

Agents can be configured with prompt files that define behavior and tool rules.

Possible prompt files:

File	Purpose
AGENTS.md	Global agent operating rules.
SOUL.md	Brand, tone, personality, and user preference context.
TOOLS.md	Tool contracts, permissions, and safety rules.
RISK.md	Trading risk rules and account limits.
SCRIPTING.md	HyperionX Code Lab scripting rules.

These files should be visible and editable by advanced users, but the platform should still enforce hard safety limits in code.

Skills

Skills are reusable instruction packs for agents.

Recommended skill layout:

~/.hyperionx/agents/skills/{skill-name}/SKILL.md

Example skills:

• Build a HyperionX indicator.
• Port an indicator from another platform.
• Review strategy risk.
• Summarize a backtest.
• Diagnose a connection error.
• Prepare a trade journal entry.
• Explain chart trader order behavior.

Tool Permissions

Agent tools should be staged by risk.

Stage	Tool Type	Risk
1	Read-only chart context	Low
2	Code Lab script draft generation	Low/medium
3	Code Lab compile diagnostics	Medium
4	Simulation-only order drafts	Medium
5	Confirmed chart trader actions	High
6	Live order submission	Highest

Live trading tools should require explicit enablement, user confirmation, risk limits, and audit logs.

Trading Settings

The advanced agent section can include live trading settings only when the product policy and entitlement allow it. Keep this section collapsed by default.

Useful fields:

• Trading enabled.
• Simulation only.
• Max order value.
• Max daily loss.
• Max position size.
• Allowed instruments.
• Allowed accounts.
• Allowed order types.
• Require confirmation before order.
• Stop trading after error.
• Stop trading after disconnection.
• Journal every action.

Strategy And Indicator Integration

An agent should be able to work with trading logic in several ways:

• Watch signals from a selected indicator.
• Watch signals from a selected strategy.
• Open Code Lab to build or edit a script.
• Create a draft strategy from a prompt.
• Explain why a signal fired.
• Review a strategy's parameters.
• Compare strategy performance across tests.

The default implementation should remain read-only and simulation-first. Live execution must stay behind product policy, entitlements, account allowlists, confirmations, risk controls, and audit logging.

Agent Builder UI

The Agent Builder window should include:

• Agent list.
• Create agent button.
• Edit agent form.
• Delete agent action.
• Provider/model selection.
• System prompt editor.
• Tool permissions.
• Memory settings.
• Workspace path.
• Skills.
• Advanced trading settings.
• API usage and cost summary.
• Performance and uptime summary.
• Recent actions/audit log.

The window should follow HyperionX dark theme rules and should not refer to third-party project names in the public UI.

Safe Build Order

Recommended implementation order:

1. Agent list and create/edit/delete UI.
2. Prompt and model settings.
3. Read-only chart context.
4. Code Lab script drafting.
5. Code Lab compile diagnostics.
6. Skills and prompt files.
7. Agent memory.
8. Backtest and optimizer result review.
9. Simulation-only actions.
10. Permissioned live trading actions after entitlement, risk, and audit review.