Everything between
your apps and the models.
The gateway handles authentication, policy, routing, caching, failover and metering, then returns one consistent response — whichever model answered.
Smart AI routing
Every request is classified before it leaves the gateway, then sent to the model that fits the job — by task type, latency budget, price ceiling or data residency.
- Policy-based rules — evaluated top to bottom, first match wins
- Canary releases — send 5% of traffic to a new model before committing
- Per-request decisions — adds 11 ms at p50, measured in the console
- Full decision trace — every route recorded with the reason it was chosen
# Routing policy — evaluated in priority order rule "regulated-data" when tag == pii route private-ai rule "complex-reason" when tokens > 8000 route claude rule "low-latency" when budget_ms < 500 route gpt-4.1 rule "bulk-classify" when batch == true route deepseek default route gpt-4.1
One integration, every model
A single OpenAI-compatible endpoint replaces every provider SDK in your codebase. Add, swap or retire a model from the console — application code never changes.
- Drop-in compatible — point your existing SDK at the gateway URL
- 40+ models — across six providers plus your own private deployments
- One key store — provider credentials never reach client code
- Streaming and tools — function calling, JSON mode and SSE pass through
# Change one line. Keep your existing SDK. client = OpenAI( base_url = "https://gateway.sparklenetworks.sg/v1", api_key = "sk-sparkle-..." ) # The gateway picks the model, applies policy and meters cost client.chat.completions.create( model = "auto", messages = [{"role": "user", "content": prompt}] )
Cost control that finance accepts
Hard and soft budgets per department, project or key. Semantic caching and tier downgrading cut spend on repeat traffic without changing what your users see.
- Department budgets — alert at threshold, throttle at the ceiling
- Semantic cache — 41% hit rate in the demo environment
- Chargeback exports — attribute every dollar to a cost centre
- Consolidated invoice — six providers, one bill, one currency
# Budget policy — Marketing department budget marketing: cap USD 500 / month warn_at 80% → email + console alert approve_at 100% → route to approval center throttle 110% → queue non-urgent traffic
Security and governance
Keys stay in the gateway. Sensitive data is redacted before egress, every request is attributable to a person, and the audit trail is immutable.
- SSO and SCIM — roles and quotas mapped from your directory
- PII redaction — detected identifiers stripped before requests leave
- Data residency lock — reject any route that would leave your region
- Immutable audit log — 7-year retention, streamed to your SIEM
# Every request carries an identity and a verdict 14:22:07 routing.rule.update K. Rahman allowed 14:08:41 failover.trigger system allowed 13:51:19 apikey.rotate M. Chen allowed 13:12:03 auth.failed unknown blocked
Built for the people
who have to sign off.
AI routing
Rules by task, latency, price and residency.
Role based access
SSO and SCIM groups map to models and quotas.
Budget control
Hard and soft limits per team, key or project.
Approval workflow
Models, prompts and keys move through review.
Automatic failover
Provider errors reroute mid-request.
Monitoring
Latency, error and token dashboards with alerting.
Billing
Chargeback by cost centre and consolidated invoicing.
Security
Redaction, injection screening, private networking.
Every feature on this page is running with simulated data in the console.
View DemoSee it routing your traffic.
A solutions engineer will walk through routing policy, cost modelling and deployment in your region.