IDENTITY & ACCESS
SSO, SCIM and role-based access
Members sign in through your identity provider over OIDC or SAML 2.0, enforced for your verified email domains. SCIM provisions and removes members and maps groups to roles.
- ✓Built-in and custom roles; every route checks a permission, not a role name
- ✓Workspaces, projects and environments that map to your teams
- ✓Access reviews with overdue escalation or automatic revocation
- ✓Scoped API keys and service accounts per project and environment
Enterprise features and availability →APPROVALS & AUDIT
Changes that need a second person
Approval policies can require a quorum before production routing or price-affecting changes take effect. Keys, policy, members and money changes are recorded in an audit trail members cannot edit or delete, and you can export it.
- ✓Policy changes preview the routes they affect before they are saved
- ✓Refunds and balance adjustments need four eyes
- ✓Staff access to your tenant is just-in-time, approved by a second person and audited
- ✓Audit export for your own retention and review
Security and staff access →BUDGETS & FINAL PRICES
Spend you can cap and explain
Budgets warn and hard-stop per key, project, workspace or organization before money moves. Every response carries its final charge, so attribution never depends on an estimate.
- ✓Each model's final price is published in the catalog: no subscription, no separate platform fee
- ✓Funds are reserved before dispatch; a failed request is not charged
- ✓Statements built line by line from rated usage
- ✓Exports that feed your own chargeback
How pricing works →ROUTING YOU CAN AUDIT
Hard constraints, ranked preferences, recorded fallback
Your policy eliminates every endpoint that breaks a region, retention, provider or capability rule, then ranks what remains by price, latency or reliability. Fallback never crosses a constraint.
- ✓Every response lists the route plan, its versions and each attempt
- ✓A request can tighten the policy, never loosen it
- ✓When nothing is eligible you get 422 no_eligible_route with the reasons, not a silent relaxation
- ✓Circuits open per endpoint and failure class; eligible traffic reroutes
Routing and fallback docs →DATA CONTROLS
Retention, training and region as rules
Request and response content is not retained by default. An organization sets a retention ceiling its projects can only tighten, can require zero-retention endpoints, and can limit processing regions.
- ✓Provider data terms are compared per endpoint in the catalog
- ✓A region is a routing constraint enforced per request, not a residency certification
- ✓Request metadata (route, usage, charge) is kept as billing and route evidence
- ✓Subprocessors are listed from the live catalog
Subprocessor list →BRING YOUR OWN KEY
Your provider keys, your wrapping key
Bring your own provider keys with an explicit scope and priority; the provider bills you directly and the credential source is recorded on every attempt. Stored credentials can be wrapped with a key your organization supplies.
- ✓Optional, labeled fallback to the shared pool
- ✓BYOK keys are encrypted at rest and never logged
- ✓External KMS references can be registered; they are not enforced
Bring your own keys docs →NETWORK & TELEMETRY
Allow-lists and telemetry to your collector
A self-serve IP allow-list restricts who can call the gateway. Request metadata, never content, can be exported to your collector over OTLP/HTTP as metrics and logs.
- ✓PrivateLink and VPC peering are requested in the console and set up by platform staff
- ✓Per-endpoint health, component status and incident subscriptions
System status →ONE API · INVOICING
One OpenAI-compatible API, billed the way you buy
Point an OpenAI SDK or an Anthropic SDK at one base URL. Chat, images, video, speech, embeddings, rerank and batches share one key, one policy and one bill.
- ✓Prepaid wallet by default, with auto top-up and low-balance alerts
- ✓Postpaid monthly statements after a credit review by platform staff, at the same final prices
- ✓Support tiers with response targets, not an SLA
SDKs and OpenAI compatibility →FINE-TUNING · OPT-IN
Fine-tuning that stays under your rules
Fine-tuning is off until an owner or security admin turns it on for the organization. Training data is sent to Together AI, the training provider, and the console says so before a job starts.
- ✓The allow_fine_tuning policy change is audited
- ✓Approval policies and budgets apply to jobs and endpoints
- ✓An organization that requires zero data retention is refused unless the provider's terms satisfy it
- ✓Fine-tuned models are private to your organization
How fine-tuning works →MCP GATEWAY
Agent tools under the same controls
Connect the MCP servers your teams use once, with managed OAuth and credentials stored encrypted, and expose only the tools you choose through toolsets that any MCP client reaches at one URL.
- ✓Registering a new server host follows your approval policies
- ✓Destructive tools wait for a person; read-only mode per toolset
- ✓Per-user connected accounts for the apps you build
- ✓Every call traced; arguments kept only if your content capture allows it
How the MCP gateway works →FEATURE AVAILABILITY · STATED AS IT IS
What is available today
Every control is available to every organization; nothing is gated by plan. What differs is how you pay and your support tier.
- SSO (OIDC and SAML 2.0)available
- SCIM provisioningavailable
- Custom rolesavailable
- Approval policiesavailable
- Regional routingavailable
- Retention controlsavailable
- Access reviewsavailable
- Audit exportavailable
- IP allow-listavailable
- Private connectivity (PrivateLink, peering)staff-fulfilled
- Customer-managed keys (customer-supplied)available
- Telemetry export (OTLP/HTTP)available
- On-prem control planeroadmap
No certification, residency claim or SLA is made: what can be obtained is listed on the security page, and support response numbers are targets.
QUESTIONS
What buyers ask first.
Write to [email protected] for postpaid billing, committed spend, an order form or the DPA.
- Which identity providers can we use?
- An identity provider that supports OIDC or SAML 2.0 signs your members in, enforced for your verified email domains. SCIM provisions and removes members and maps groups to roles.
- Is Yelno SOC 2 or ISO 27001 certified?
- No. No SOC 2 examination has been completed and no ISO/IEC 27001 certification is held. The data processing addendum and the subprocessor list are available on request; the security page lists what can be obtained.
- Is there an SLA?
- No contractual SLA is offered. Support tiers carry response targets, and they are labeled as targets.
- Can we pay by invoice?
- Yes, after a credit review. Postpaid organizations are settled by a monthly statement at the same final prices; it is enabled by platform staff, not by a self-serve switch.
- Can we use our own provider keys?
- Yes. Bring-your-own-key credentials carry an explicit scope and priority, the provider bills you directly, and every attempt records which credential served it.
- Where does request content go?
- Only to the provider endpoints your policy allows. Content is not retained by default, retention and region are hard routing constraints, and an endpoint that breaks one is eliminated with a stated reason.