SentinelX for Somnia

Keep Somnia oracles honest, resilient, and operator-controlled

SentinelX gives protocol teams a reliable safety net: real-time monitors, AI-assisted incidents, and an allowlisted execution plane that keeps humans in control while the system does the heavy lifting.

  • Guardians approve every action while GuardianHub enforces allowlists.
  • Telemetry flows into Convex so drift, staleness, and remediation are auditable.
  • AI triage writes actionable incidents even when external LLMs are offline.

Why teams choose SentinelX

Autonomy when you want it, human control when you need it. SentinelX anchors incident response in clear policy, verifiable data, and accountable execution.

• Autonomous monitors catch drift across Chainlink, DIA, and custom feeds before slippage hits pools.

• AI generates severity, mitigations, and playbooks, while operators stay in charge of the final transaction.

• Convex persistence and GuardianHub guardrails make audits and post-incident reviews straightforward.

The reliability gap on Somnia today

Volatility, thin liquidity, and multi-source oracle stacks make reliability a full-time job. Without automated guardrails, teams respond after losses. SentinelX starts by confronting the root issues.

Reality lags the oracle heartbeat

Price feeds wait for deviation thresholds or heartbeat timers before updating. During Somnia volatility, teams trade on stale numbers for minutes.

Chainlink data feeds

Multi-source configs drift apart

Somnia DIA feeds recommend 0.5% deviation and 120s freshness; Chainlink defaults to 1% / 180s. Without continuous alignment, governance votes lock in bad assumptions.

DIA price feeds on Somnia

Operational mistakes cascade fast

Paused keeper jobs, MEV manipulations, and manual incident response leave users exposed. SRE-grade telemetry, AI triage, and gated execution are now table stakes.

Flashbots research

How it works (end‑to‑end)

A layman’s walkthrough from install to safe execution on Somnia.

Install & register

Install SentinelX and register your Somnia contract (vault, marketplace, etc.) in the web app.

Wire pause hook

Paste a tiny mixin or inherit a base so your contract can be paused by SentinelX if risk is detected.

Read two feeds

SentinelX reads two on-chain price feeds (Protofire/Chainlink + DIA) for the same asset and marks Unsafe if they drift or go stale.

AI policy agent

Simple rules + LLM summary decide whether to pause via the on-chain guardian and draft mitigations.

Operator control

Dashboard shows incident timelines and decisions; operators can one‑click unpause when conditions recover.

SafePrice output

Optionally expose a SafePrice view (median/guarded) in your app to reduce reliance on a single oracle.

The SentinelX stack keeps coverage, diagnosis, and action in sync

Every layer is purpose-built: monitors gather truth, AI provides clarity, and the control plane makes mitigations safe to execute. Each piece can run on its own, but SentinelX shines when everything is connected.

Data plane

Policy runner evaluates SafeOracleRouter on schedule, compares Chainlink, DIA, and custom feeds, and stores telemetry snapshots in Convex.

Every monitor keeps historical drift, stale durations, and operator overrides for post-mortems.

AI plane

LLM summaries translate anomalies into human context: severity, root cause, mitigations, and fallback logic when OpenAI is unreachable.

Operators can see both the raw data and the AI judgement to decide faster.

Control plane

GuardianHub + AgentInbox hold allowlisted contract calls. Operators approve or reject action intents and relay mitigations with tracked Somnia transaction hashes.

All activity rides through audit-friendly pipelines and sticks to allowlists.

Response playbook

SentinelX keeps the loop tight—from detection to mitigation—without removing human oversight.

Watch continuously

SafeOracleRouter monitors run every cycle, comparing multi-source data and contract thresholds so drift never slips through nightly cron gaps.

Diagnose with context

Telemetry is normalized into incidents with AI-written context plus deterministic fallbacks. Teams see what moved, why it matters, and how long it has been broken.

Decide and act safely

Action intents craft GuardianHub calldata, preflight simulations, and execution plans. Operators approve the single allowlisted tx, keeping final accountability.

Why not “just integrate an oracle”?

Teams already use oracles. SentinelX handles what surrounds them: guardrails, monitoring, triage, and safe execution.

Deviation + heartbeat means expected staleness

Feeds update on deviation thresholds or on a heartbeat—staleness and drift are normal operating states, not edge cases (see Chainlink model).

Somnia DIA defaults are concrete

DIA recommends ~0.5% deviation and 120s refresh with a 24h heartbeat; a price can be “valid” yet stale up to the heartbeat.

Manipulation and MEV pressure exists

Flash-loan manipulation and MEV can skew edge conditions. Circuit breakers and guardians exist to buy time in real incidents.

What SentinelX does that a single oracle doesn’t

Concrete guardrails and a production runbook—wired for Somnia.

Multi-source guardrails

Compare DIA + Protofire/Chainlink and enforce deviation/heartbeat health; open incidents when feeds disagree or go stale.

Data → AI → Control

Data plane writes telemetry/incidents, AI plane drafts summaries and Action Intents, control plane executes with human approval and tx hashes.

Time-to-respond value

Ships a production runbook in a box for Somnia—cut MTTD/MTTR without rebuilding monitors, approvals, and guardian flows from scratch.

Somnia-first defaults

Pre-filled addresses and DIA’s update model baked in; infra settings are read-only by default with advanced overrides for power users.

Auditability & safety

Every propose → approve → execute step is recorded; allowlists ensure only intended calldata reaches chain.

Buy vs Build

The job-to-be-done comparison for Somnia teams.

Detect stale / divergent feeds
Just integrate oracleYou must build it
SentinelXBuilt-in monitors (deviation/heartbeat, cross-feed)
Explain incidents fast
Just integrate oracleManual triage
SentinelXAI summaries (severity, root cause, mitigations)
Decide safe next step
Just integrate oracleAd-hoc runbooks
SentinelXStructured Action Intents (pause/tune params)
Execute safely on-chain
Just integrate oracleWrite your own guardian infra
SentinelXGuardianHub/AgentInbox + approvals + tx recording
Prove you handled it
Just integrate oracleManual notes
SentinelXAudit trail (who approved, which tx)

On Somnia, SentinelX is the reliability layer above your oracles—detecting drift/staleness across sources, auto-drafting safe responses with AI, and letting your team approve a one‑click on‑chain fix.

See how it’s wired

Operational outcomes SentinelX delivers

Measure success in minutes saved, incidents contained, and mitigations that land safely the first time.

24 / 7

Coverage windows

Monitoring never pauses—policy runner keeps tenants online across weekends, holidays, and volatile Somnia trading sessions.

< 60s

Incident triage

AI summaries land in the dashboard within a minute of drift detection so operators can contain issues before they hit liquidity pools.

1 tx

Controlled execution

Allowlisted AgentInbox relays mean mitigations reach chain with a single signed transaction and full audit trails.

Built for production teams

Everything you need to ship reliable price-driven agents and dApps on Somnia—from cron-friendly indexers and action queues to docs that cover REST hooks, webhooks, and policy runners.

Multi-source oracle guardrails

Set deviation, quorum, and heartbeat per asset. Compare Chainlink, DIA, or custom feeds and surface incidents before end users feel them.

AI triage & action intents

Incident summaries, mitigations, and GuardianHub calldata arrive in seconds so operators focus on decisions, not copy-paste playbooks.

Production-ready integration

RainbowKit SIWE auth, Convex persistence, cron-ready policy runner, and docs with API/webhook examples make SentinelX deployable on day one.

Ready to guard your protocol?

Spin up a tenant, seed demo data, and run the full SentinelX loop locally. When you are ready for mainnet, configure the GuardianHub and AgentInbox addresses once and ship with confidence.