In-App Agent UI / UX

Dynamic Context Adaptation

User context is not treated as static. Recommendations are adjusted dynamically based on observable system and market-level signals, combined with the user’s selected profile parameters.

No persistent behavioral profiling or autonomous decision-making is performed.

To support transparency, the adaptation process follows a clearly defined workflow that governs how signals are detected, evaluated, and surfaced to the user.

Agent Workflow Overview

The Agent operates as a context-aware orchestration layer, responsible for evaluating external signals and preparing informational outputs. It does not execute transactions or alter user positions.

The following diagram illustrates the Agent’s high-level workflow, from signal intake to user-visible outcomes.

Market and Protocol Signals (Examples):

The system processes a range of predefined signals. Each signal triggers a scoped system response and, where relevant, a user-facing message.

Network conditions

• Ethereum transaction fees exceed predefined thresholds

  → System response: Adjust relative network weighting in visible recommendations

  → User message: "Ethereum transaction fees are currently elevated. Alternative networks with lower execution costs are available."

Protocol risk events

• A supported protocol reports a security incident

  → System response: Remove protocol from active recommendations

  → User message: "Protocol X has been excluded from recommendations following a reported security incident."

Protocol lifecycle updates

• A new protocol completes a predefined vetting period

  → System response: Make protocol eligible for profile-based inclusion

  → User message: "Protocol Y has completed the vetting process and is now eligible for inclusion based on profile parameters."

Performance metric changes

• Reported APY for a previously visible protocol decreases materially

  → System response: Surface updated data and comparable metrics

  → User message:: "The reported APY for Protocol X has decreased to 2.2%. Updated yield data for comparable protocols is available."

The Continuous Adaptation Loop

Adaptation is implemented as a continuous, event-driven loop, ensuring that surfaced information remains aligned with current conditions.

This loop enables timely updates while preserving full user control at every stage.

In-App Agent Chat

The in-app Agent chat serves as an interactive access layer to the adaptation and recommendation system described above. It allows users to query, explore, and contextualize information already available within the application, without transferring decision-making authority to the Agent.

Purpose and Scope

The Agent chat is designed to:

reduce interface friction

improve discoverability of relevant information

provide contextual explanations of system outputs

It does not function as an autonomous advisor, portfolio manager, or execution engine.

All responses are informational and scoped to:

application state

protocol visibility

user-selected preferences and constraints

Predefined Prompts and Guided Queries

To support usability, the chat interface includes a set of predefined prompts that reflect common user intents, such as:

“How can I deploy 10,000 in the current setup?”

“How can I review my current allocations?”

“Why was this protocol surfaced?”

“What changed since my last interaction?”

These prompts are templates rather than instructions. The Agent responds by explaining available options, relevant constraints, and visible data, without recommending specific actions.

Context-Aware UI Guidance

When a user asks questions related to application navigation or available actions, the Agent may provide visual guidance within the interface.

This can include:

highlighting relevant sections

pointing to specific controls or filters

guiding the user through multi-step flows

This guidance is non-executive:

no buttons are pressed automatically

no transactions are prepared or submitted without user initiation

The goal is to help users understand where actions can be taken, not which actions should be taken.

Agent Chat Workflow

The following diagram illustrates how user queries are processed by the in-app Agent chat and translated into contextual explanations and interface guidance.

Decision Authority and Control

The Agent chat does not:

initiate transactions

alter allocations

override user preferences

perform actions on behalf of the user

Any operational step — including transaction preparation — requires an explicit user action outside the chat interface and a cryptographic signature.