> ## Documentation Index
> Fetch the complete documentation index at: https://developers.octaviatech.app/llms.txt
> Use this file to discover all available pages before exploring further.
# Build with Octavia
> Which service to start with, and how to put together a real integration.
This page is a decision guide. It answers two questions: *which service do I activate
first*, and *how do I actually wire one up* so it survives contact with production.
***
## Start here
Building editorial content, a blog, or a multi-language site? Start with
**AI CMS** — it is the service with a full reference today. Storing user uploads
or generated media? **Storage**. Authenticating your own users? **Auth**.
Taking money? **Payments**.
What each service covers.
Enable the service from **Dashboard → Octavia Services**, then choose a plan. Every
plan is the same API — the difference is what you are allowed to do before the API
answers 426.
The click-by-click version, with screenshots.
One key per system. Start with Read-only for anything that only
reads, and raise the role only when a call actually fails with
403.
Roles, scopes, and rotation.
Official clients exist for JavaScript, Python, PHP, C# and Go. They send the key
for you, unwrap the envelope, and give you typed results.
Compare the five and pick one.
***
## Pick an SDK, not a language
The five SDKs wrap the same REST endpoints and return the same envelope. They differ
only in how the language spells things.
| **If your stack is** | **Use** | **Why** |
| --------------------------- | ---------- | --------------------------------------------------------------------------------- |
| Node, Next.js, React | JavaScript | TypeScript-first, covers the full surface, and is the one we build against first. |
| Django, FastAPI, scripts | Python | No third-party dependencies. |
| Laravel, WordPress, Symfony | PHP | Uses cURL directly, so there is no PSR-18 client to wire up. |
| ASP.NET, Unity, Xamarin | C# | .NET 6+, returns `CMSResponse` for every method. |
| Go microservices | Go | A single module, zero dependencies outside the standard library. |
Every SDK sends exactly one header, `x-api-key`, and none of them expose a way to
add headers. If you find yourself wanting to set a header manually, the answer is in
the gateway, not the client.
***
## Put together something real
For AI CMS, the content model has an order. Building it in this order means each step
needs only what is above it.
```mermaid theme={null}
flowchart LR
L["1 · Language"] --> A["2 · Author"]
L --> C["2 · Category"]
C --> SC["3 · Subcategory"]
A --> ART["4 · Article"]
C --> ART
SC --> ART
```
Every resource, its fields, and what it depends on.
Build the whole model in about five minutes.
***
## Before you ship
Four things separate a working demo from an integration that holds up.
**Handle `429` and `426` differently.** `429` is transient — back off exponentially
with jitter. `426` means a plan quota is exhausted, and retrying cannot fix it. Putting
them in the same retry loop just burns rate-limit budget.
**Set a timeout.** The PHP SDK defaults to *no timeout*, so a hung request will hang
your worker. Set `timeoutMs` explicitly in every language.
Which failures you can retry, and which you cannot.
What each plan includes, limit by limit.
**Keep keys on the server.** A key in a browser bundle is a compromised key. Read it
from the environment, give each system its own, and rotate on a schedule.
**Paginate from the start.** `total` counts the filtered set, not the page. Build the
loop now rather than after the first timeout.
***
Every code and its meaning.
Ask the engineering team.