Skip to main content
The Sentinel Core is the central nervous system of the SURCHI protocol. It is the coordination layer that receives data from the pipeline, routes intelligence tasks to the appropriate AI agents, manages cross-agent communication, enforces risk parameters, and delivers outputs to the Natural Language Interface and execution layer. No sentinel operates in isolation — every signal and every action passes through the Sentinel Core, which ensures coherence, corroboration, and compliance with user-defined constraints before anything reaches the chain.

​
Architecture Overview

The Sentinel Core sits between the Data Pipeline and the three AI sentinels, acting as the orchestration layer that makes the system function as a unified intelligence engine rather than a collection of independent agents. Its core responsibilities are:
1

Data Reception

The Sentinel Core receives preprocessed, normalized feature vectors from the Data Pipeline. Raw on-chain events, DEX order book updates, liquidity metrics, and sentiment signals have already been normalized and tagged with metadata before reaching the Core. The Core does not perform raw data processing — it operates on structured intelligence inputs.
2

Task Routing

Based on signal type and current protocol state, the Core routes intelligence tasks to the appropriate sentinel. Price movement signals and whale wallet activity route to Alpha Sentinel. Liquidity depth changes and pool health metrics route to Liquidity Sentinel. Approved strategy conditions route to Execution Sentinel. Routing is deterministic and auditable.
3

Cross-Signal Correlation

Some market events generate correlated signals across multiple sentinel domains. The Core manages cross-sentinel communication — sharing relevant outputs between agents so that, for example, a liquidity deterioration signal from Liquidity Sentinel can inform the risk calculation of an active Execution Sentinel strategy. No sentinel operates with a siloed view of the market.
4

Risk Parameter Enforcement

Before any output reaches the Execution Sentinel, the Core validates it against the user’s approved risk parameters. Position limits, slippage tolerances, stop-loss conditions, and circuit breaker states are all evaluated at the Core level, not delegated to individual sentinels. This ensures a single, consistent enforcement point.
5

NLI Output Publication

Intelligence outputs — market signals, liquidity alerts, execution confirmations — are formatted and published to the Natural Language Interface for delivery to the user. The Core handles translation from structured signal objects to human-readable intelligence summaries.

​
The Three Sentinels

Alpha Sentinel

Market intelligence: price analysis, whale tracking, cross-market signal synthesis, and opportunity identification.

Liquidity Sentinel

Liquidity protection: pool health monitoring, impermanent loss analysis, and liquidity risk early warning.

Execution Sentinel

Autonomous on-chain execution: strategy execution, transaction submission, and position management within approved parameters.

​
Orchestration Model

The Sentinel Core uses a consensus-based orchestration model for high-stakes execution decisions. For routine signals — alerts, informational outputs, low-value transactions — a single sentinel’s output is sufficient to proceed. For execution events that exceed defined value thresholds or involve compounding risk across multiple domains, the Core requires corroboration from multiple sentinels before proceeding. For example: if the Execution Sentinel is prepared to execute a large position entry, but the Liquidity Sentinel has simultaneously flagged deteriorating pool depth in the target market, the Core holds the execution and presents the conflict to the user for resolution. The sentinels do not override each other — the Core surfaces the conflict and defers to user intent. Failure handling is built into the orchestration model. If a sentinel fails to return a signal within its latency SLA, the Core applies a conservative fallback: it does not guess, and it does not proceed with incomplete information. Active strategies are paused, and users are notified. Recovery procedures are documented and testable.

​
On-Chain Transparency

Every significant decision made by the Sentinel Core is recorded as an on-chain event on Solana. This includes:
  • Signal routing decisions and the data inputs that triggered them
  • Risk parameter evaluations and the outcomes of those checks
  • Execution approvals and rejections, with reasoning
  • Circuit breaker activations and deactivations
  • Cross-sentinel corroboration events
These records are immutable, timestamped, and publicly verifiable. Any participant — user, auditor, or researcher — can reconstruct the decision history of the protocol from on-chain data alone. Transparency is not a dashboard feature; it is a protocol-level guarantee.
The Sentinel Core is non-custodial. It never holds user funds — it only submits signed transactions on behalf of the user within approved parameters.