Resilience control planeDemo gateway active

Live DemoDemo gateway

Dashboard / Resilience Control Plane

Agent Control Plane

Monitor failover, MCP health, gateway policy, and graceful user experience during infrastructure chaos.

Resilience Score

policy-weighted continuity

Fallbacks

automatic recovery paths

Retries

budgeted attempts this run

Incidents

degraded events contained

Incident Commander

Submit the user-facing request and observe the recovery path.

ready

request idreq_pending

gateway routetfy-gateway/us-east/primary-gpt

confidence92%

Live Agent Health

Model mesh, gateway, and MCP fleet status.

Model meshcontinuityops/incident-agent

success118ms

AI Gatewaytfy-gateway/us-east

success99.95%

MCP Gateway5 governed tool servers

success5 servers

Recovery Analytics

Fallback activity across the current incident run

Weekly activity

Average: 62%Peak: 90%

Route Distribution

Where the agent found continuity

Primary routes5

Recovered routes2

Degraded routes1

Human approval0

Resilience Timeline

Every gateway, model, policy, and MCP step in order.

Primary request accepted08:48:24 AM

req_9df3a21 routed through tfy-gateway/us-east

TrueFoundry AI Gateway14ms

Provider instability detected08:48:25 AM

Claude unavailable, routed request through fallback policy

TrueFoundry AI Gateway610mspreview_recovery_path

MCP evidence recovered08:48:26 AM

Metrics server returned 500, switched to cached tool response

TrueFoundry MCP Gateway900mspreview_recovery_path

User response preserved08:48:27 AM

Gateway selected backup model based on latency and reliability score

TrueFoundry AI Gateway142mspreview_recovery_path

Recovery Path

Model fallback, retry policy, and user continuity

armed

Primaryopenai/gpt-4.1healthy

Gateway policytfy-gateway/us-east/primary-gptreliability score 0.94

User responsepreserved2.4s recovery

Gateway will preserve the user response through degraded mode.

When a provider, gateway route, or MCP server fails, ContinuityOps records the failure, applies retry and fallback policy, and returns a transparent response instead of a dead-end error.

Chaos Lab

Inject realistic infrastructure failures.

LLM timeoutPrimary model exceeds response budgetopenai/gpt-4.1Claude unavailableProvider route returns repeated 5xxanthropic/claudeRate limitProvider APIs throttle callsprovider.statusGateway brownoutPrimary route ejected by policytfy-gateway/us-eastMCP crashMetrics server returns 500metrics.queryInvalid tool responseLog search sends malformed JSONlogs.searchWrite deniedTicket MCP requires approvalticket.createPartial outageRunbook data is incompleterunbook.lookup

MCP Server Fleet

Tool status, latency, and recovery action.

MCP serverStatusLatency

Metricsmetrics.query

success84ms

Prometheus adapter healthy

Logslogs.search

success116ms

Log search index current

Runbooksrunbook.lookup

success42ms

Cached runbooks warmed

Statuspagestatuspage.check

success64ms

Provider status reachable

Ticketingticket.create

success95ms

Write path authorized

User Experience Continuity

The response the operator receives after degraded-mode recovery.

systemReady

Run the demo to show how the user receives a useful response even when the model route or MCP server fails.