SaaS and AI vendors
Your customers, your agents, your models: each in its own lane.
Flexway helps you serve several tenants, govern your customers' agents and keep the cost of your models in check, from your own infrastructure.
What Flexway decides, on your infrastructure
Who calls, what Flexway decides, what is called: each route is a use case of your sector. Dashed: what is coming.
Architecture for a SaaS or AI vendor: your customers' apps, agents and integrators go through Flexway, on your infrastructure. Flexway checks identity and tenant, offers MCP tools derived from the contract, holds each tenant's inference budget and quota, in front of your services, your MCP servers, your internal models and model APIs. A tenant over its quota gets a 429. Coming: third-party agents (A2A) and the semantic cache.
- Customer apps
- The applications that call your APIs: each request goes through Flexway, which checks the contract and the identity before reaching your systems.
- Customer agents
- AI agents: they call MCP tools and models through Flexway, under the same identity, quota and audit as an API.
- Integrators
- The integrators who connect your customers: API key or client certificate bound to a tenant, under that tenant's quota.
- Third-party agents
- Under way: Third-party agents talking to your agents (A2A): signed agent cards, traffic under quota and audit.
- MCP tools from the contract
- The MCP tools come from your contract and your federated MCP servers, filtered by scope, with bounded delegation.
- Identity and tenant
- Flexway checks who is calling (OIDC token, API key or client certificate) and reads the tenant from the identity; no valid identity: 401.
- Tenant quota
- The tenant's daily and monthly quota applies across all its routes; beyond it: 429 quota_exceeded.
- Semantic cache
- Planned: A cache of inference responses by similarity of meaning, enabled operation by operation.
- Inference budget per tenant
- The tenant's budget, in tokens and euros, is reserved before the model call; exhausted: 429 budget_exceeded.
- Multi-tenant services
- Your REST and gRPC services: they receive only the requests Flexway let through, under deadline and circuit breaker.
- Existing MCP servers
- Your existing MCP servers, federated behind Flexway: their tools come under the same authorization, quota and audit.
- Internal models
- Your internal models, called through the OpenAI-compatible surface; a sensitive request is pinned to them.
- Model APIs
- The AI providers outside your infrastructure: their keys stay in Flexway, each call counts against the tenant's budget.
- Line: one route
- Interchange: shared decision
- Request
- Model or tool call
- Refusal
- Your infrastructure
- Planned
- MCP tools from the contractscopes, bounded delegation
- Identity and tenantAPI key, mTLS or OIDC
- Tenant quotadaily and monthly, shedding
- Semantic cacheplanned
- Inference budget per tenantdeclared euros, ordered fallback
1A customer plugs their agent into your product
What Flexway decidesMCP tools come from your contract; the agent acts with bounded scopes and delegation.
What is protectedYour services only see declared operations; every call keeps an identity and a cost.
2An AI feature of your product calls a model
What Flexway decidesThe tenant's euro budget is checked, the model chosen by policy, with an ordered fallback.
What is protectedYour model bill stays capped, tenant by tenant.
3A large customer floods the API
What Flexway decidesIts tenant's daily and monthly quota applies: beyond it, a 429; under saturation, load shedding.
What is protectedThe other tenants keep their throughput.
Customer apps
The applications that call your APIs: each request goes through Flexway, which checks the contract and the identity before reaching your systems.
Routes through here
Customer agents
AI agents: they call MCP tools and models through Flexway, under the same identity, quota and audit as an API.
Routes through here
Integrators
The integrators who connect your customers: API key or client certificate bound to a tenant, under that tenant's quota.
Routes through here
Third-party agents
Under way: Third-party agents talking to your agents (A2A): signed agent cards, traffic under quota and audit.
- Catalogue features
- A2A governance: signed agent cards, traffic under quota and audit
MCP tools from the contract
The MCP tools come from your contract and your federated MCP servers, filtered by scope, with bounded delegation.
- Decision
- scopes, bounded delegation
- Catalogue features
- MCP surface: the contract's operations offered as tools ; Delegation bounded in time and depth, cascading revocation ; MCP federation: existing MCP servers under the same authorisation, quota and audit
Routes through here
Identity and tenant
Flexway checks who is calling (OIDC token, API key or client certificate) and reads the tenant from the identity; no valid identity: 401.
- Decision
- API key, mTLS or OIDC
- Catalogue features
- API keys and client mTLS as first-class identities ; iss and aud checked; several issuers in one configuration
Routes through here
Tenant quota
The tenant's daily and monthly quota applies across all its routes; beyond it: 429 quota_exceeded.
- Decision
- daily and monthly, shedding
- Catalogue features
- Quota per tenant per day and month ; Adaptive load shedding under saturation
Routes through here
Semantic cache
Planned: A cache of inference responses by similarity of meaning, enabled operation by operation.
- Catalogue features
- Semantic cache, per operation
Inference budget per tenant
The tenant's budget, in tokens and euros, is reserved before the model call; exhausted: 429 budget_exceeded.
- Decision
- declared euros, ordered fallback
- Catalogue features
- Budget and rate per tenant ; Policy routing across providers and models
Routes through here
Multi-tenant services
Your REST and gRPC services: they receive only the requests Flexway let through, under deadline and circuit breaker.
Routes through here
Existing MCP servers
Your existing MCP servers, federated behind Flexway: their tools come under the same authorization, quota and audit.
- Catalogue features
- MCP federation: existing MCP servers under the same authorisation, quota and audit
Routes through here
Internal models
Your internal models, called through the OpenAI-compatible surface; a sensitive request is pinned to them.
- Catalogue features
- Backends by kind (OpenAI, relay, gRPC), local or upstream metering
Routes through here
Model APIs
The AI providers outside your infrastructure: their keys stay in Flexway, each call counts against the tenant's budget.
- Catalogue features
- Policy routing across providers and models
Routes through here
What the sector requires
- Multi-tenantOne tenant never spills over another
- Daily and monthly quotas per tenant, a cache isolated per tenant and subject, load shedding when the load rises.
- AgentsYour customers' agents, under control
- MCP tools from your contract, a bounded agent chain, governed A2A: every call keeps an identity, scopes and a cost.
- AI costA predictable model bill
- Budgets in euros at declared prices, a soft cap, an ordered fallback to another model before the outage.
- IntegrationsYour integrators, authenticated
- API keys and client certificates bound to a tenant and to operations, OIDC for your users.
What the sector expects
The expectations of your sector found in the market analysis, and where Flexway stands on each; what another component does better is stated.
| What enterprises expect | Flexway v1.6.3 |
|---|---|
| API gateway in front of React, Next.js, mobile, microservices | Delivered · since v0.1.0 |
| BFF: adapt and aggregate several microservicesBody projection built; aggregation needs a trade-off decision; per-screen composition stays with the customer's BFF | Partly · next in 1.10 |
| Rights per user, tenant, application, resourceUser, tenant, application built; the right on the object (BOLA) stays with the service | Partly · next to be scheduled |
| GraphQL: limit query complexityProposed decision: declared persisted queries, whose cost is judged at registration | To be scheduled |
| Several API styles behind one entry point | Delivered · since v1.1.0 |
| Distinct policies for public, partner and internal APIs | Delivered · since v0.1.0 |
| AI-augmented frontend with no provider key, streams, limits | Delivered · since v0.1.0 |
| Follow the AI calls of a featureUse case declared in the contract; a session is followed through the trace | Partly · next in 1.8 |
| Event and AI gateways brought together | Delivered · since v1.5.0 |
| Gateway driven by business intentThe declared intent (use case, task route) is resolved by a rule; never by a model | Partly · next in 1.9 |
| Interactions between agentsGoverned A2A built, signed cards and registry planned | Partly · next in 1.9 |
| Self-optimising gatewayRecommendations as reviewed proposals; autonomous application excluded | Partly · next to be scheduled |
Upcoming releases are not commitments; "Delivered" only when the catalogue of the published release says so.
How Flexway answers
One tenant never spills over another
- Quota per tenant per day and monthDelivered · v0.1.0
- Adaptive load shedding under saturationDelivered · v0.1.0
- Authenticated cache, isolated per tenant and subjectDelivered · v1.1.0
- Rate per caller (token bucket)Delivered · v0.1.0
- In-flight request cap per upstream (bulkhead)Delivered · v0.1.0
- Circuit breaker with half-open stateDelivered · v0.1.0
Your customers' agents, under control
- MCP surface: the contract's operations offered as toolsDelivered · v0.1.0
- MCP federation: existing MCP servers under the same authorisation, quota and auditDelivered · v0.1.0
- Delegation bounded in time and depth, cascading revocationDelivered · v0.1.0
- A2A governance: signed agent cards, traffic under quota and auditIn progress
A predictable model bill
- OpenAI-compatible /v1/* surfaceDelivered · v0.1.0
- Budget and rate per tenantDelivered · v0.1.0
- Sensitive requests pinned, sovereignty and fallback by policyDelivered · v0.1.0
- Backends by kind (OpenAI, relay, gRPC), local or upstream meteringDelivered · v1.5.0
- Semantic cache, per operationPlanned
Your integrators, authenticated
- API keys and client mTLS as first-class identitiesDelivered · v0.1.0
- iss and aud checked; several issuers in one configurationDelivered · v0.1.0
- An OpenAPI 3.1 contract loaded; what is served is derived from itDelivered · v0.1.0
State in Flexway's latest published release; a feature is marked delivered only with its release.
What Flexway does not do
- Flexway does not bill your customers: it measures tokens and euros, your billing uses them.
- Per-user budgets and the semantic cache are planned, not delivered yet: their version is shown.
Let's talk about your gateway
Flexwiz is a young company: you talk directly with the person who founds it. Tell us what you want to run through Flexway. We talk about your infrastructure and your digital sovereignty requirements.
Online booking is coming soon; until then, an e-mail is enough.