Voice Pipeline Architecture
Every millisecond matters in voice AI. PulseGuard's modular pipeline lets you bring your own models — swap any stage to match your latency and accuracy requirements:
Pipeline Stage
Your Choice
Target Latency
Voice Activity Detection
Configurable VAD engine
< 50ms
Speech-to-Text
Any Whisper-compatible STT
< 150ms
LLM Reasoning
Any OpenAI-compatible model
< 150ms
Text-to-Speech
Any streaming TTS engine
< 100ms
Total Voice-to-Voice
Optimized pipeline
< 400ms
Next-Gen: Unified Omni Models
PulseGuard supports emerging omni-models that handle speech-in → speech-out in a single model pass, reducing the pipeline to one stage. With the right hardware, target latency drops to ~200ms voice-to-voice.
Telephony Integration
PulseGuard supports multiple telephony providers for maximum flexibility:
FreeSWITCH Twilio Telnyx Vonage SIP Trunking Inbound IVR
Service Map & Blast Radius
PulseGuard builds a complete service topology of your infrastructure. When an incident occurs, it instantly calculates the blast radius — mapping every downstream service affected by the failure.
Service Dependencies
Track four dependency types: runtime, build, data, and async. Each with latency metrics and protocol specifications.
Priority Tiers
Classify services P0-P5 with automatic escalation rules based on tier. P0 services trigger immediate voice calls to the on-call team.
RED Metrics
Rate, Error, and Duration metrics per service node. Drill down from service → traces → spans for full observability.
Endpoint Monitoring
PulseGuard continuously monitors your services with configurable health checks:
HTTP Checks TCP Checks PING Checks DNS Checks 10-3600s Intervals
Alert Pipeline
PulseGuard's alert pipeline is designed for zero-noise, high-signal incident detection. Every alert goes through deduplication, enrichment, and intelligent routing before anyone gets notified.
Webhook Ingestion
HMAC-SHA256 signed webhooks from Prometheus, Grafana, Zabbix, or custom JSON. Pre-built parsers extract severity, fingerprint, and context automatically.
Smart Deduplication
Fingerprint-based dedup with configurable windows (default 5 min). Occurrence tracking counts how many times each alert fired. No more alert storms.
AI-Powered Enrichment
Each alert is enriched with AI context per webhook source, service map linkage, and root cause suggestions via LLM analysis of related alerts and service context.
Escalation Engine
Multi-step escalation policies with configurable delays and repeat loops. Target resolution across Primary, Secondary, and Manager on-call layers.
Alert Lifecycle
State
Description
Actions Available
OPEN
New alert, awaiting response
Acknowledge, Snooze, Resolve
ACKNOWLEDGED
Engineer is investigating
Resolve, Escalate, Add Note
SNOOZED
Temporarily muted
Unsnooze, Resolve
RESOLVED
Incident closed
Reopen, View Timeline
Notification Channels
Four notification methods with intelligent dispatch:
Smart Voice Dispatch
Voice calls are blocked if the tenant already has an active call — the system falls back to push notifications as a bridge. Configurable concurrency limits (default: 2 concurrent voice calls per tenant) with 1-hour TTL and automatic requeue.
On-Call & Scheduling
PulseGuard's on-call system supports the complexity of real engineering organizations with multi-layer rotations and timezone-aware scheduling.
3-Layer Rotations
Primary, Secondary, and Manager escalation layers. Each layer has its own rotation schedule with configurable handoff times and rotation periods.
Override Management
Temporary overrides for vacations, sick days, or shift swaps. Override windows with start/end times and automatic reversion to the base schedule.
Timezone Support
Follow-the-sun rotations across global teams. Each schedule respects local timezones for handoff times, ensuring engineers are always paged during reasonable hours.
Automation & MCP Tools
PulseGuard's automation engine lets you define conditional triggers that execute remediation actions automatically — or through the Voice AI during a call.
MCP Tool Types
SSH Execution
Execute commands on remote servers directly through the AI. Run diagnostics, restart services, check logs — the AI interprets results and reports back.
REST API Calls
Call any HTTP endpoint: Kubernetes API for pod management, webhook triggers for CI/CD pipelines, or custom internal APIs. Full request/response logging.
Database Queries
Run SQL queries to diagnose issues: check connection pools, find slow queries, verify replication status. Read-only mode available for safety.
Ticket Creation
Automatically create Jira tickets with full incident context: timeline, affected services, blast radius, and remediation steps taken. Bi-directional sync.
MCP Server Management
Connect and manage MCP servers with enterprise-grade reliability:
HTTP Transport WebSocket Transport STDIO Transport Vault-Backed Secrets Circuit Breaker Auto-Discovery 30s Timeout Health Monitoring
Circuit Breaker Protection
Every MCP server connection is protected by a circuit breaker (CLOSED → OPEN → HALF_OPEN). If a tool server becomes unresponsive, PulseGuard automatically stops sending requests and attempts gradual recovery — preventing cascade failures in your automation pipeline.
Automation Rules
Define rules that trigger automatically based on alert conditions:
- Conditional Triggers — Fire based on severity, service, labels, or custom conditions
- Tool Chains — Execute multiple MCP tools in sequence with result passing
- Test Mode — Validate rules in a sandbox before deploying to production
- Execution Tracking — Every rule execution logged with SUCCESS, FAILURE, or SKIPPED status
Multi-Tenancy & Security
PulseGuard is built from the ground up as a multi-tenant SaaS platform with enterprise-grade security at every layer.
Tenant Isolation
Complete data isolation per tenant. All database queries, queue operations, and API calls are namespaced by tenantId. No data leakage between organizations.
Vault-Backed Secrets
All credentials, API keys, and MCP server secrets are stored in HashiCorp Vault — never in the database. Automatic rotation and audit logging for every secret access.
RBAC
Three built-in roles: Admin, Operator, and Viewer. Fine-grained permissions control who can acknowledge alerts, modify schedules, execute tools, or manage integrations.
Zitadel Authentication
Enterprise SSO via Zitadel with support for OIDC, SAML, and social logins. Multi-factor authentication and session management with configurable policies.
Audit & Compliance
- Full Audit Trail — Every action logged: alert acknowledgments, tool executions, schedule changes, configuration updates
- Voice Call Recording — Complete transcripts and recordings of all AI voice calls for compliance and training
- API Access Logging — All 100+ API endpoints log request metadata with tenant context
- Incident Timeline — Detailed timeline for every incident showing escalation steps, notifications, and resolution actions
Architecture at a Glance
AI Model Allocation
Purpose
Configuration
Context
Voice Conversations
Any OpenAI-compatible LLM
Configurable context window
Alert Summarization
Configurable per tenant
LLM analysis
Root Cause Reasoning
Configurable per tenant
Related alerts + service context
Semantic Embedding
Configurable per tenant
Alert dedup + similarity
Load Balancing Strategies
PulseGuard supports four load balancing strategies across AI model providers:
PRIORITY ROUND_ROBIN REGION_BASED COST_OPTIMIZED
With ordered fallback chains — if the primary model provider goes down, PulseGuard automatically fails over to the next provider in the chain with zero downtime.
Why Teams Choose PulseGuard
OpsGenie Migration
With Atlassian sunsetting OpsGenie, teams need a modern alternative. PulseGuard offers a smooth migration path with API compatibility and feature parity — plus Voice AI that OpsGenie never had.
Self-Hosted Option
Deploy PulseGuard in your own infrastructure for complete data sovereignty. All AI models can run on-premises via Ollama, vLLM, or LocalAI. No data ever leaves your network.
Developer-First API
100+ REST API endpoints with comprehensive documentation. Everything in the UI can be automated via API. Webhook-first architecture for seamless integration with your existing tools.
Ready to modernize your incident management?
PulseGuard transforms incident response from a manual, stressful process into an AI-assisted, automated workflow. Your engineers sleep better. Your customers see fewer outages. Your MTTR drops from hours to seconds.
Book a demo →