Each application holds its own provider key, policy and logs.
AISIX AI Gateway
AISIX is an AI-native gateway built for model and agent traffic. It gives every application one controlled way to reach models: stable model aliases, central credentials, routing, limits, guardrails and usage telemetry in a single control plane.
One gateway · many models · centrally controlled. AISIX is part of the DOME CLOUD AI infrastructure practice. It is not Apache APISIX or API7: AISIX handles AI traffic, while APISIX and API7 handle general API traffic.
AI is already in the company. The control layer usually is not.
Most teams adopted AI faster than the organisation could govern it. AISIX sits between your applications and AI providers, so every request is authenticated, routed, limited and recorded in one place.
Applications call aliases. AISIX holds the keys, applies policy and records usage.
Five control problems AISIX closes
Fragmented credentials
Every team stores its own provider keys and settings.
Central caller keys and provider credentials, scoped by model and tool.
Vendor lock-in
Changing a model means changing every application.
Stable model aliases, so applications stop hard-coding providers.
Uncontrolled spend
Token usage and provider bills are spread across teams.
Usage and budget controls per caller, where the edition supports it.
Inconsistent policies
Each application has its own limits, guardrails and logs.
One policy plane for traffic that passes through the gateway.
Limited visibility
Requests, latency, errors and AI usage are scattered.
Usage events, metrics and audit history in one place.
You probably need AISIX if...
Most companies answer yes to at least two.
Book a readiness assessment- 01 Your teams already use more than one AI provider.
- 02 More than one application holds its own provider key.
- 03 You need AI spend visibility by team or application.
- 04 You need consistent limits and guardrails.
- 05 Agents or tools can take actions with AI.
- 06 You need audit evidence for AI usage.
Two or more yes? Start with an AISIX readiness assessment.
Three services, delivered end to end
Scope it precisely, deliver it completely.
Access & credentials
Caller keys, model and tool permissions, and provider credentials held centrally instead of inside every application.
Model routing
Stable model aliases with routing and failover, so you can change providers without changing application code.
Policy & visibility
Rate limits, budgets and guardrails per caller, with usage events, metrics and audit history across AI traffic. Budget controls depend on edition.
See one request move through AISIX.
Switch between the four controls to see what AISIX does before a model provider is called. This is a simulation, not live traffic.
A stable name the app calls. The provider behind it can change.
- Authenticate
- Authorize
- Apply controls
- Route
- Observe
- Return
The application never holds a provider key.
- App key verified
- Provider keys stay inside AISIX
A stable name the app calls. The provider behind it can change.
- Authenticate
- Authorize
- Apply controls
- Route
- Observe
- Return
The application still calls the same alias.
- The alias does not change
A stable name the app calls. The provider behind it can change.
- Authenticate
- Authorize
- Apply controls
- Route
- Observe
- Return
Policy is enforced before the provider is called.
- Guardrail check passed
A stable name the app calls. The provider behind it can change.
- Authenticate
- Authorize
- Apply controls
- Route
- Observe
- Return
Usage, errors and policy decisions are visible.
- hr-analytics · summary-standard 1,240 โทเคน · สำเร็จ
- finance-insights · analysis-standard 860 โทเคน · สำเร็จ
- coding-agent · code-standard 2,410 โทเคน · ถูกปฏิเสธตามนโยบาย
- crm · summary-standard 640 โทเคน · สำเร็จ
One gateway. Three ways to run the control plane.
In every edition the gateway runtime stays in your environment. The difference is where the control plane lives and how much management is built in.
AISIX OSS
Open sourceTeams that operate the gateway themselves.
- Gateway runtime
- Customer-operated
- Control plane
- Not included
- Configuration
- Declarative / self-managed
- Usage, budgets & audit
- Added separately
AISIX Enterprise
Self-hosted control planeOrganisations that need to run the control plane themselves.
- Gateway runtime
- Customer-operated
- Control plane
- Customer-hosted
- Configuration
- Dashboard + Admin API
- Usage, budgets & audit
- Built in, by edition
AISIX Cloud
Hosted control planeTeams that want a hosted control plane.
- Gateway runtime
- Customer-operated
- Control plane
- API7-hosted
- Configuration
- Dashboard + Admin API
- Usage, budgets & audit
- Built in, by edition
Availability, edition and commercial terms are confirmed during the assessment.
Not sure which edition fits?
Start with a short assessment, or prove it with a PoC against a real use case.
AI traffic is different.
AISIX sits beside your existing API gateway. One controls AI and agent traffic; the other keeps handling general APIs.
Traditional API Gateway
HTTP / APIAuth and routing for REST, GraphQL and internal services, with plugins, rate limits and service management.
- HTTP / API traffic
- Plugins and rate limits
- REST, GraphQL, internal services
AISIX AI Gateway
AI / agentsA gateway purpose-built for model and agent traffic: model aliases, provider credentials, AI-specific routing and telemetry.
- AI / agent traffic
- Model aliases and provider keys
- Guardrails, MCP / A2A where supported
A gateway controls traffic. Governance covers the organisation.
AISIX enforces policy on every AI request that passes through it. Identity, data, cost and incident response still need to connect around it.
AISIX enforces
- Caller access
- Model aliases and routing
- Limits and guardrails
- Usage telemetry
IAM & access
Joiner, mover, leaver and service identities
Data & privacy
Classification, DLP and application security
AI FinOps
Cost allocation and procurement
SIEM & response
Audit retention, detection and incident response
Resilience
Application resilience, recovery and BCP
AISIX makes AI traffic visible. The surrounding controls make it accountable.
A 90-day path, not a big-bang rollout.
Start with one or two real use cases. Prove control before scaling the programme.
Assess and design
Inventory AI usage and risk, map data and identity requirements, and choose pilot applications.
- AI usage and risk inventory
- Target architecture
- PoC plan
Prove it
Stand up AISIX, connect one or two applications and providers, and test allowed, denied and failure paths.
- AISIX environment
- Verified PoC
- Decision log
Govern and scale
Write policies and operating procedures, add budget and audit controls, train the team, and plan the production rollout.
- AI policies
- Budget and audit controls
- Production rollout plan
Built to hold up in production
One control plane
Applications stop holding provider keys and per-app policies.
Switch models safely
Move traffic between providers at the alias, not in code.
See the spend
Token and usage visibility by caller, team or application, by edition.
Built for agents
MCP tools and A2A agents connect through the gateway where supported.
AISIX ≠ Apache APISIX ≠ API7
AISIX handles AI and agent traffic; Apache APISIX and API7 handle general API traffic. They are designed to run side by side, connected through shared identity and policy.
Ready to raise the ceiling on your platform?
Tell us what you're building. We'll assess it and propose the right approach, and the first conversation is free.
Messenger · m.me/domecloud.io