Priority
Send through the preferred provider. Fall back only on failure or saturation.
Routing engine
One routing brain for every event, every channel.
Pick a strategy, score providers, branch on tenant tier, fail over when a vendor goes down, retry with backoff. All configured in the admin, never coded into your applications.
Live routing
Events / min12,478+4.2%
Sources
API
/notifications
Kafka
events.notif
Webhook
ingest
Routing engine
Trigger
Health
Strategy
Providers
Twilio
412ms
SendGrid
298ms
FCM
1.2s
Slack
190ms
Strategy · priority · per-tenantFCM degraded · failing over to APNs
Pick how the next provider gets chosen.
The strategy is per-channel, per-tenant. Change it in the admin and the next event picks up the new rule. No redeploy, no migration, no application restart.
- Strategy per channel, per tenant, per environment.
- Live changes take effect on the next event.
- Per-provider scoring weights you can dial in.
- Audit log of every strategy change with the actor.
- Combine multiple strategies into a pipeline.
Round-robin
Even rotation across N providers when no preference applies.
Region
Geo-aware per locale. Route MA traffic through INWI / MGate, EU through Twilio.
Tenant tier
Premium tenants get the low-latency provider pool; standard tenants get the cost-optimised one.
SMS chain · tenant acme-prod
live- PrimaryHealthy
Twilio
SMS · low-latency
- SecondaryHealthy
MGate
SMS · regional fallback
- TertiaryHealthy
Vonage
SMS · cost-optimised
Vendor down? The chain takes over.
Primary, secondary, tertiary, all per channel and per tenant. Health probes pull an unhealthy provider out of rotation before the next event picks it. When the primary recovers, traffic shifts back without an operator step.
- Health probes every 5 seconds, per provider.
- Chain configured in the admin, not in code.
- Automatic swap back when the primary stabilises.
- Per-channel chains so SMS and email failover separately.
- Manual override available from the admin or API.
Transient failures retry. Without melting the queue.
Exponential backoff with jitter, capped at a per-channel ceiling. A flapping provider gets fewer attempts as the gap grows, so one bad vendor never takes the whole pipeline down with it.
- Exponential backoff, configurable per channel.
- Jitter prevents thundering-herd retry storms.
- Max attempts capped per event class.
- Dead-letter queue for events that exhaust the chain.
- Per-event retry metrics in the audit trail.
Exponential backoff
base 1s · max 32s · jitter ±20%1stry 1
2stry 2
4stry 3
8stry 4
16stry 5
32stry 6
Routing rules · order-events
3 rules · first-match- 1IF
tenant.tier == 'premium' AND channel == 'sms'Matched→ strategy: priority · pool: low-latency
- 2ELSE IF
payload.region IN ('MA','TN','DZ')→ strategy: region · pool: MA-providers
- 3ELSE
*→ strategy: round-robin · pool: default
Branch on the things that actually change.
Tenant tier, channel, payload fields, time of day. Branches are first-match. Order them in the admin once and the audit trail explains every routing decision back to the rule that won.
- First-match branching, ordered visually.
- Conditions on payload, tenant, channel, time.
- Every match recorded with the rule ID and timestamp.
- Test a rule against historical events before publishing.
- Per-tenant rule sets, version-controlled.
Every routing decision, written down.
For every event, the audit trail records the strategy that ran, the rule that matched, the provider chosen, and the outcome. When someone asks “why did this notification go that way?”, you point at a row.
- Per-event timeline: chosen, sent, retried, succeeded.
- Filter by tenant, channel, provider, outcome.
- p95 latency per provider, live.
- Counter of events using the fallback in the last hour.
- Export the trail to your warehouse on a schedule.
Routing decisions · live
streaming- 10:42:18.412
ORDER_SHIPPED· acme-prod
→ Twiliovia priority
412mssent - 10:42:18.407
PASSWORD_RESET· atlas
→ SendGridvia rule#1
298mssent - 10:42:18.401
ORDER_DELAYED· northwind
→ MGate (fallback)via failover
612mssent - 10:42:18.395
OTP_REQUEST· acme-prod
→ Twiliovia priority
388mssent - 10:42:18.387
MARKETING_BLAST· northwind
→ Mailgunvia round-robin
1184msretry
More platform features
Keep exploring.
TemplatesHandlebars-style syntax with conditionals, loops, and helpers. Multi-language.Multi-tenancyRepository-level isolation per tenant. Per-tenant providers, audit, quotas.Audit trailEvery event and every admin action, per tenant. PII redacted at the field.Quotas & rate limitsPer-tenant, per-event thresholds with warning and critical alerts.ObservabilityEvents per minute, delivery rate, p95 latency, per-provider health, live.Agent CLIThe Nemofy skill installs in Claude Code, Codex, Cursor, OpenCode, Aider.