Platform
Bring your own LLM for serious AI software delivery
Your provider keys, OpenAI-compatible endpoints and private model routing on enterprise plans — the full delivery loop, running on models your organization has approved.
Bring your own LLM lets enterprises run Ciao on their own model terms: your provider keys, OpenAI-compatible endpoints — including privately hosted models — and private model routing on enterprise plans. Unlike platforms that force one bundled model arrangement, Ciao keeps the delivery loop — Builder, QA, Security, Guardrails — while inference runs under agreements and boundaries your security and legal teams already approved.
Published 2026-07-03 · Last updated 2026-07-03
Your AI platform should run on your model terms
For many enterprises, the blocker to AI-assisted engineering is not capability — it is model governance. Legal has negotiated terms with specific providers. Security maintains an approved list. Some workloads must not cross a private boundary at all. A platform that bundles its own model arrangement and asks you to accept it forces a choice between your policies and your delivery speed.
Ciao removes that choice. Bring your own LLM: use your own provider keys so inference runs under agreements you negotiated, point routing at OpenAI-compatible endpoints — including models you host privately — and configure private model routing on enterprise plans. The delivery loop stays; the models answer to you.
This is not an edge case. In most large organizations the model question is settled before the platform question is even asked — and the platforms that survive review are the ones that fit the settled answer rather than trying to reopen it.
How bring-your-own-LLM works
Configuration is administrative and audited — set once by the people accountable for it, inherited by everyone who builds.
1. Bring your provider keys
Inference runs against your accounts, under the commercial and data terms your organization already negotiated with the provider.
2. Or point at OpenAI-compatible endpoints
Any endpoint speaking the OpenAI-compatible API can serve the platform — including models hosted inside your own boundary.
3. Configure private model routing
Decide which models handle which work across your workspaces. Enterprise plans make routing an administrative decision, not a per-prompt negotiation.
4. Keep ladder discipline
Fallback and task routing still apply within the set you approve — resilience without stepping outside your boundaries. The failure of one approved endpoint does not stall delivery when a second is configured.
5. Govern it like everything else
SSO via SAML and OIDC, optional MFA and RBAC control access; configuration and admin actions land in the append-only audit trail.
Why it matters
Own-model support turns the hardest procurement conversations into short ones. Data flows to providers you chose, under contracts you negotiated, inside boundaries you drew — so the AI platform review stops being a new legal negotiation and becomes an extension of decisions already made.
It also pairs naturally with the rest of an enterprise posture: private VPC or on-prem deployment for the applications, custom sandboxes for existing stacks, and private model routing for inference. Each piece answers to the same principle — your software, your infrastructure, your models.
And it keeps optionality where it belongs. Providers will keep changing; hosting economics will keep shifting. When routing is yours to configure, those shifts become administrative updates rather than platform migrations.
Who brings their own models
Own-model routing is typically driven by the people who sign the risk register:
- CISOs and security architects — Inference constrained to approved providers or private endpoints, verifiable in configuration rather than promised in slides.
- Enterprises with negotiated agreements — Existing provider contracts and committed spend put to work instead of duplicated by a platform bundle.
- Regulated organizations — Financial services, healthcare operations and government teams whose workloads must stay inside a private boundary.
- Platform teams — One sanctioned model gateway serving every AI initiative, with Ciao routing through it like everything else.
Security and governance notes
The model configuration itself is governed:
- ✓ Own provider keys, OpenAI-compatible endpoints and private model routing are available on enterprise plans.
- ✓ Where Ciao-managed models are used, inference runs under zero-retention model contracts.
- ✓ Customer code is not used to train models — under any configuration.
- ✓ SSO via SAML and OIDC, optional MFA and role-based access control govern administration.
- ✓ Model configuration changes and admin actions land in the append-only audit trail.
- ✓ SOC 2 Type II reports are available under NDA.
Model options on Ciao
Four options that compose rather than compete:
| Option | What it is | Typical fit |
|---|---|---|
| Ciao-managed model ladder | Multi-provider routing with fallback, under zero-retention contracts | Teams that want results without managing model vendors |
| Your provider keys | Inference through your own accounts and negotiated terms | Enterprises with existing provider agreements and committed spend |
| OpenAI-compatible endpoint | Any compatible endpoint, including privately hosted models | Organizations standardizing on a model gateway or private hosting |
| Private model routing | Administrative control over which models handle which work | Regulated teams with approved-model policies per workload |
Frequently asked questions
Can we use models we host ourselves?
Yes. Routing can point at OpenAI-compatible endpoints, which includes models hosted inside your own environment. Combined with private VPC or on-prem deployment for the applications, the whole loop can run within boundaries you control.
Does any of our code go to providers we have not approved?
Own-model configurations exist precisely to prevent that: routing is constrained to your keys and endpoints, and configuration changes are recorded in the append-only audit trail. Where Ciao-managed models are used instead, inference runs under zero-retention contracts and customer code is not used to train models.
Which plans include bring-your-own-LLM?
Own provider keys, OpenAI-compatible endpoints and private model routing are enterprise-plan capabilities. Serious production programs start at USD 10,000 per year — talk to sales about the configuration your security team needs.
Do we lose the model ladder if we bring our own models?
No. Task routing and fallback discipline apply within the set of models you approve, so you keep the resilience of a ladder without stepping outside your governance boundaries.
How does this fit with custom sandboxes and on-prem deployment?
They compose. Custom sandboxes bring the delivery loop to your existing Rails, Java, Go, Python, Node or multi-process stack; private VPC and on-prem deployment (under separate terms) keep the applications inside your boundary; own-model routing does the same for inference.