Agent Builder
Agent Builder creates specialized Rion profiles for jobs such as chart analysis, Code Lab assistance, strategy research, validation, and audit review.
Agent Builder requires the agent_builder entitlement and a Pro, Elite, or Admin license. Profile editing and selected context controls are implemented, but the full risk, telemetry, packaging, and public automation contract is not complete.
Open Agent Builder
Open Agent Builder from the main HyperionX window or an available Rion entry point. If the signed-in license does not include the entitlement, HyperionX displays a locked message instead of opening the window.
The window participates in workspace restoration, but an agent's own file workspace is separate from the desktop window layout.
Create A Specialized Profile
- Create or select a profile.
- Give it one narrow role, such as chart analysis, compiler repair, Validator review, or Research Lab review.
- Select an enabled AI provider and model.
- Write system instructions that state what evidence the agent must use and what it must not claim.
- Choose the context and tool toggles described under Enforced Profile Controls.
- Select a workspace root and review every prompt or skill file in it.
- Treat the remaining risk and automation fields as advisory unless this page identifies them as enforced.
- Save the profile, select it in the supported Rion workflow, and test with read-only requests first.
A profile does not install a new model provider. Provider credentials remain in the global AI settings and are not copied into the profile.
Workspace Root
The default workspace is:
%USERPROFILE%\Documents\HyperionX\Agents\Workspace
Skills normally live below the selected workspace:
%USERPROFILE%\Documents\HyperionX\Agents\Workspace\skills\<skill>\SKILL.md
Changing a workspace path does not create an operating-system sandbox. Keep the profile inside a dedicated, reviewable directory and do not point it at a home directory, source monorepo, credential directory, or synced folder containing unrelated private data.
Generated Prompt Files
The current builder can create or manage this prompt-document set:
| File | Intended use |
|---|---|
README.md | Workspace overview. |
AGENTS.md | Operating rules and task boundaries. |
SOUL.md | Voice, tone, and behavioral context. |
TOOLS.md | Tool guidance and safety rules. |
IDENTITY.md | Agent role and identity. |
USER.md | User-specific working preferences. |
BOOTSTRAP.md | Startup context for the workspace. |
RISK.md and SCRIPTING.md are not generated by the current builder. You may create additional documentation for human use, but its presence does not make a rule a hard runtime control.
Enforced Profile Controls
For a non-baseline specialized profile selected by the in-app Rion workflow, the execution policy currently enforces these controls:
| Control | Runtime effect |
|---|---|
| Chart and chart-data access | Can hide symbol/timeframe and chart counts when disabled. |
| Market-data access | Can remove recent and visible candles. |
| Indicator/script read access | Can remove indicator context and restrict Code Lab prompt context. |
| Chart Trader state | Can replace Chart Trader details with a restricted snapshot. |
| Strategy/tool state | Can hide strategy, Validator, Optimizer, and Research Lab state. |
| File access | Participates in blocking Code Lab/script context and mutations for that selected profile. It is not an OS filesystem sandbox. |
| Order routing | Blocks a recognized in-app trading action for that selected profile. Global product gates still apply. |
| Read-only default | Blocks recognized mutating tool actions. |
| Block market orders | Blocks recognized market-order placement for that profile. |
Multiple toggles can participate in one permission. For example, indicator context requires both the general indicator/script permission and indicator read permission.
The built-in profiles whose IDs are agent and rion-scout currently use the unrestricted baseline execution policy. Editing their displayed permission toggles does not apply the specialized-profile filter described above. Use a non-baseline specialized profile when you need those filters, and keep all global simulation and entitlement gates enabled.
Advisory And Not-Yet-Enforced Fields
The builder stores additional planning and risk fields, but the current profile policy does not enforce them as hard limits:
- maximum order quantity or value
- daily loss stop
- maximum position size
- maximum trades per day
- allowed account, instrument, or order-type text
- stop-after-error and stop-on-disconnect instructions
- require-approval and trading-approval fields
- journal/audit preference fields
- heartbeat, queue, routing, and automation descriptions
- stop-loss, take-profit, buy-condition, and sell-condition text
These values can guide the model through profile instructions. They do not independently inspect the account or block an order. Enforce real trading limits in the account, strategy, simulation environment, and supported execution controls.
Relationship To The Local Desktop API
Agent Builder permissions apply to selected in-app specialized-agent execution. They do not create a separate Local Desktop API token, and they do not filter the global HTTP API's snapshots or permissions.
The Preview HTTP API separately checks its token, endpoint permission, license entitlements, local settings, and—only for trading—the Hyperion Sim account restriction. See Local Desktop API.
Trading Boundary
The current Local Desktop API supports simulated trading actions only. Agent Builder fields do not enable live HTTP order routing.
Before any simulated action:
- verify the selected account is Hyperion Sim
- review symbol, side, type, price, and quantity
- require a user decision unless the supported simulation auto-execute policy was deliberately enabled
- inspect the returned action result and account state
Do not present a profile's text fields as platform-enforced risk management.
Import And Export
Agent Builder currently uses different formats in each direction:
- Import accepts a JSON agent profile.
- Export creates a
HyperionX.AgentBuilder.Exportversion 1 ZIP archive.
An export can contain:
agent.json- a manifest
- prompt documents
- selected skill-directory files
- selected indicator or strategy source
The builder does not currently import its own ZIP export. Treat the ZIP as a handoff and review archive, not as an installable package.
Before exporting:
- keep the workspace and skills inside a narrow dedicated root
- inspect the generated manifest and archive contents
- remove customer data, local paths, secrets, and unrelated source
- verify the selected script source is intended for distribution
Before importing JSON, review provider, endpoint, workspace, skill, and permission fields. Agent profiles are trusted local configuration and are not signed Marketplace packages.
Telemetry And Audit Display
API usage/cost, uptime, performance, trade count, and recent-action summaries shown in the current Agent Builder UI are placeholders. They are not live billing, uptime, or trading telemetry.
Use provider billing, HyperionX account/performance tools, and the Local Desktop API audit files as their documented source of truth. Audit files record requested actions; they do not prove an exchange or broker fill.
Storage And Sharing
Profiles are stored as readable JSON at:
%APPDATA%\HyperionX\AIAgent\agents.json
Provider API keys are stored separately in protected AI settings and are excluded from profile exports. Profile prompts, paths, skills, and bundled source can still be sensitive. See AI Data and Privacy.