> For the complete documentation index, see [llms.txt](https://candora.gitbook.io/whitepaper/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://candora.gitbook.io/whitepaper/candora-pro/candora-edge/infrastructure.md). # Infrastructure ### Agent Runtime, Enforcement, and Automation Control Layer Edge Infrastructure is the operational foundation of [Candora Edge](/whitepaper/candora-pro/candora-edge.md). Its purpose is to provide a controlled environment in which AI agents can observe markets, generate decisions, and participate in trading activity while remaining fully subject to the same rules that govern every participant on the exchange. While [Candora Agent](/whitepaper/candora-pro/candora-edge/candora-agent.md) provides Candora's native trading assistant and Event Intelligence capabilities, [Agent Builder](/whitepaper/candora-pro/candora-edge/agent-builder.md) focuses on creating custom automation, and [Edge Marketplace](/whitepaper/candora-pro/candora-edge/marketplace.md) enables distribution and monetization, Edge Infrastructure serves as the runtime, enforcement, permission-control, and operational-safety framework that supports the entire ecosystem. The framework exists to solve a fundamental challenge in automated trading systems: allowing software to participate in markets without allowing software to bypass the rules that govern those markets. Many automation platforms combine observation, decision-making, execution, and access control into a single operational layer. While this can simplify deployment, it also concentrates authority inside systems that may become difficult to inspect, constrain, or disable when market conditions deteriorate. Candora takes a different approach. Edge Infrastructure separates observation, intelligence generation, action formation, execution preparation, and enforcement into independent layers. This separation allows automation to remain powerful while ensuring that market authority remains with the exchange rather than with the software operating inside it. The result is an automation framework designed around accountability, deterministic behavior, operational safety, and controlled market participation. ### Architectural model Edge Infrastructure is organized as a layered runtime architecture that separates information gathering, decision generation, execution preparation, and enforcement. At the foundation of the architecture is the Observation Layer, which collects approved inputs including market-state information, liquidity conditions, account constraints, execution telemetry, and selected external information sources. These inputs are consumed by the Intelligence Layer, where agent logic transforms information into signals, recommendations, forecasts, and trading intent. Once intent has been generated, the Execution Layer converts that intent into structured requests capable of being evaluated against Candora's risk controls, market-integrity systems, and admission requirements. Before any action reaches the market, it must pass through the Enforcement Layer, where Permission Envelopes, platform safeguards, and exchange-level protections are applied. This separation ensures that analytical systems remain distinct from market authority. Agents may generate ideas, recommendations, and actions, but they cannot independently redefine how the exchange processes those actions. ### Enforcement principles Edge Infrastructure is built around a set of architectural principles that remain true regardless of agent type, strategy complexity, runtime configuration, or market conditions. Observation does not grant execution authority. Reasoning does not grant execution authority. Agent-generated intent cannot bypass exchange controls. Permission Envelopes cannot expand automatically. Agent actions remain subject to the same execution rules as all other participants. Exchange enforcement remains independent of agent-generated decisions. These principles establish a clear separation between intelligence generation and market authority throughout the Candora ecosystem. An agent may form an opinion, generate a recommendation, or prepare an action for execution. It may not alter the rules of the market, bypass platform controls, or obtain authority beyond the permissions explicitly granted to it. ### Agent lifecycle Every Edge agent begins operation with a predefined runtime profile and a participant-defined Permission Envelope. As market conditions evolve, the agent continuously observes approved inputs and evaluates those inputs against its strategy logic, operating policies, and participant-defined constraints. When predefined conditions are satisfied, the agent may generate a bounded action request. Before that request can reach the market, it is validated against all applicable controls, including Permission Envelope restrictions, platform risk requirements, market-integrity protections, and exchange admission rules. Only after successful validation may an action enter Candora's standard execution path. Whether an action is executed or rejected, the resulting decision becomes part of the agent's permanent audit history. At no stage does an agent receive direct access to matching, settlement, valuation, custody, or market-state infrastructure. Logic remains upstream while authority remains with the exchange. ### Controlled input framework Edge agents may operate using a broad range of approved information sources, including market data, volatility conditions, liquidity metrics, account constraints, execution activity, and external intelligence feeds. Access to information does not imply unrestricted access to infrastructure. Agents operate exclusively through approved interfaces and bounded data sources. They cannot establish arbitrary network connections, inspect protected services, access hidden infrastructure state, or consume information beyond their assigned scope. This design preserves operational isolation while ensuring that agent behavior remains reproducible, inspectable, and suitable for deterministic replay. ### Permission Envelopes Every Edge agent operates within a Permission Envelope established by the participant. The Permission Envelope defines the markets an agent may access, the actions it may perform, the amount of capital it may deploy, the exposure it may assume, and the conditions under which it may operate. These permissions function as hard constraints rather than advisory guidelines. Actions that exceed permitted limits are automatically rejected before execution. Agents cannot modify their own permissions, expand their authority, or grant themselves access to additional functionality. Authority therefore remains explicitly controlled by the participant rather than by the automation itself. ### Runtime isolation and sandboxing Edge Infrastructure treats operational failures as a first-class design concern. Agents execute within isolated runtime environments governed by restrictions on memory consumption, processor utilization, execution duration, connectivity, action frequency, and permitted system operations. These controls are designed to prevent software failures from becoming infrastructure failures. An Edge agent cannot access [Orbit](/whitepaper/candora-grid/orbit.md) directly, modify [Vault](/whitepaper/candora-grid/vault.md) records, bypass [Risk Engine](/whitepaper/candora-grid/risk-engine.md) controls, or establish alternative execution pathways. Sandboxing ensures that software remains a participant within the platform rather than becoming part of the platform itself. ### Failure handling Edge Infrastructure is designed to fail safely. When an agent loses access to required inputs, exceeds runtime limits, generates malformed output, violates permission constraints, or encounters enforcement controls, authority contracts rather than expands. Depending on the nature of the failure, the agent may be paused, restricted, reset, or quarantined. Actions that cannot be validated are rejected before execution. This behavior is intentional. During abnormal market conditions, uncertainty should reduce automation authority rather than increase it. The platform therefore defaults toward controlled contraction whenever confidence in an agent's operating state deteriorates. ### Auditability and replay Every material decision generated by an Edge agent produces a structured audit record. These records capture observed inputs, active permissions, generated intent, validation outcomes, execution requests, and resulting actions associated with a decision. The resulting audit trail supports operational review, historical reconstruction, simulation, forensic analysis, testing, and performance evaluation. Edge follows the same deterministic event-ordering principles used throughout [Candora Grid](/whitepaper/candora-grid.md). Equivalent inputs, permissions, and operating conditions therefore reconstruct equivalent decision histories during replay. This allows automation behavior to be evaluated independently from market outcomes and provides a consistent foundation for verification, debugging, and accountability. ### Multi-agent coordination Edge Infrastructure supports the deployment of multiple cooperating agents operating within a shared strategy framework. Different agents may specialize in market observation, event analysis, liquidity interpretation, portfolio supervision, risk monitoring, or execution coordination. These agents may share information and coordinate activity while remaining independently permissioned, individually auditable, and fully subject to the same enforcement framework as any standalone system. Coordination increases analytical coverage and operational flexibility without creating privileged authority or reducing accountability. ### Relationship to Candora Grid Edge Infrastructure operates alongside the broader Candora Grid architecture while remaining subordinate to its enforcement systems. [Orbit](/whitepaper/candora-grid/orbit.md) remains responsible for execution. [Oracle](/whitepaper/candora-grid/oracle.md) remains responsible for valuation. [Pulse](/whitepaper/candora-grid/pulse.md) remains responsible for market-state synchronization. [Risk Engine](/whitepaper/candora-grid/risk-engine.md) remains responsible for solvency controls. [Vault](/whitepaper/candora-grid/vault.md) remains responsible for settlement finality and ownership records. Edge Infrastructure does not replace these systems. Instead, it provides a controlled environment in which automation can observe, reason, and act while remaining fully subject to the rules that govern the exchange. ### System role Edge Infrastructure serves as the runtime foundation of Candora Edge. It provides the agent execution environment, permission-control framework, validation systems, sandboxing architecture, audit infrastructure, replay capabilities, and operational safety mechanisms that allow intelligent automation to function without compromising market integrity. By separating intelligence generation from market authority, Edge Infrastructure enables sophisticated automation while preserving the fairness, determinism, transparency, and operational integrity of the Candora ecosystem.