Developers

Developer guide. OpenAI and Anthropic SDKs, curl and Claude Code.

Situra speaks both the OpenAI and the Anthropic formats. Point your SDK, framework or coding agent at Situra’s URL, use an situ_ key, and you are done.

Gateway URLhttps://api.situra.ai

Quickstart. From zero to your first request.

Create an account

In the Situra console, with your work email. Test keys (situ_test_) work straight away, before your organisation is verified.
Create a project

Choose its residency tier — ES, EU or Global — and logging mode. Residency applies to every key in the project.

ESEUGlobal
Generate a key

With an expiry and, optionally, a model allowlist. The secret is shown once: put it in your secret manager.
Change the base URL

In your SDK or agent. The rest of your code stays as it is.

Model names in the samples are illustrative. GET /v1/models returns the models your project may use, given its tier.

Frameworks such as LangChain, LlamaIndex, the Vercel AI SDK or LiteLLM connect through their OpenAI-compatible client, set to this base URL and key. We do not maintain framework-specific integrations.

 app.py  
 import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.situra.ai/v1",
    api_key=os.environ["SITURA_API_KEY"],  # situ_live_…
)

response = client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "Summarise this case file in three sentences."}],
)
print(response.choices[0].message.content) 

 app.mjs  
 import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.situra.ai/v1",
  apiKey: process.env.SITURA_API_KEY, // situ_live_…
});

const stream = await client.chat.completions.create({
  model: "gpt-5-mini",
  messages: [{ role: "user", content: "Summarise this case file in three sentences." }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
} 

 app.py  
 import os
from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.situra.ai",
    api_key=os.environ["SITURA_API_KEY"],  # situ_live_…
)

message = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarise this case file in three sentences."}],
)
print(message.content[0].text) 

 app.ts  
 import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://api.situra.ai",
  apiKey: process.env.SITURA_API_KEY, // situ_live_…
});

const message = await client.messages.create({
  model: "claude-sonnet-4-6",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Summarise this case file in three sentences." }],
});
console.log(message.content); 

 Terminal  
 curl -i https://api.situra.ai/v1/chat/completions \
  -H "Authorization: Bearer $SITURA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "x-situra-residency: es" \
  -d '{
    "model": "gpt-5-mini",
    "messages": [{"role": "user", "content": "Summarise this case file in three sentences."}]
  }'

# Response headers:
# x-situra-residency: es
# x-situra-route-region: spaincentral
# x-situra-attempts: 1 

The x-situra-residency header is optional and can only tighten the project’s tier, never loosen it.

 Terminal  
 # Point Claude Code at Situra
export ANTHROPIC_BASE_URL="https://api.situra.ai"
export ANTHROPIC_AUTH_TOKEN="situ_live_…"
# Model IDs from Situra’s catalogue (dated IDs do not exist)
export ANTHROPIC_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-4-6"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-haiku-4-5"

claude 

Claude Code sends the key as Authorization: Bearer; Situra also accepts x-api-key. Pin the models with the ANTHROPIC_*_MODEL variables: by default Claude Code asks for dated IDs (claude-haiku-4-5-20251001) that Situra does not recognise. The models must be allowed in the project, and the project’s tier decides where they run.

Console playground. Try models before writing code.

The console playground sends real requests through the gateway with one of your keys and shows the residency, route, latency and cost of every response.

Situra console playground — Real console with demo data.

Coding agents. Claude Code and other agents, with the same governance.

Coding agents send a lot of context: source code, paths, sometimes secrets. Routing them through Situra applies the project’s residency, budgets and metadata logging just like any other application. Tools that accept an OpenAI-compatible base URL work the same way.

One key per team or developer, with an expiry
A monthly budget with a hard stop
EU or ES residency for your source code
Per-key usage in the console

 ~/.zshrc  
 export ANTHROPIC_BASE_URL="https://api.situra.ai"
export ANTHROPIC_AUTH_TOKEN="situ_live_…"   # key for the “agents” project
export ANTHROPIC_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-4-6"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-4-6"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-haiku-4-5" 

Anthropic format with curl

The /v1/messages endpoint accepts the key in x-api-key, like Anthropic’s API.

 Terminal  
 curl https://api.situra.ai/v1/messages \
  -H "x-api-key: $SITURA_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hola"}]
  }' 

Test keys. Wire up CI before you are verified.

situ_test_ keys are served by a sandbox built into the gateway: deterministic responses, realistic token counts and zero cost. They work even while your organisation is pending verification, so you can connect your tests on day one.

 Terminal  
 export SITURA_API_KEY="situ_test_…"

# The same code as before. With a test key:
# - the built-in sandbox answers, deterministically
# - with realistic token counts
# - zero cost, and records flagged as sandbox
python app.py 

API surface. Endpoints, headers and errors.

Endpoints
Endpoints
POST	`/v1/chat/completions`	OpenAI
POST	`/v1/embeddings`	OpenAI
GET	`/v1/models`	OpenAI
POST	`/v1/messages`	Anthropic
POST	`/v1/messages/count_tokens`	Anthropic
GET	`/.well-known/situra-attestation-keys`	JWKS

Response headers
Response headers
`x-request-id`	Request identifier, the same one you will see in the console
`x-situra-residency`	Tier of the route that served the request: the project’s tier or a stricter one. Only on responses a route served (not on 502 or 503)
`x-situra-route-region`	Region of that route, under the same condition
`x-situra-attempts`	Number of routes tried
`x-situra-attestation`	Signed attestation (EdDSA JWT) of where the request was processed; verify it with the keys at `/.well-known/situra-attestation-keys`
`x-ratelimit-limit-requests / -remaining-requests`	Per-minute request limit (the tighter of the key’s and the organisation’s) and what is left
`x-ratelimit-limit-tokens / -remaining-tokens`	The organisation’s per-minute token limit and what is left
`retry-after`	Seconds to wait before retrying (429 and some 503s)

Status codes

Status codes
HTTP	Code (error.code)	Meaning
`401`	`missing_api_key`, `invalid_api_key`, `expired_api_key`	Missing, invalid, expired or revoked key
`402`	`insufficient_balance`, `credit_limit_reached`, `budget_exceeded`	Insufficient balance (prepaid), credit limit reached (invoiced), or a hard-stop budget is exhausted
`403`	`organization_pending_verification`	Your organisation is pending verification: until then only test keys (`situ_test_`) work
`403`	`model_not_allowed`, `organization_suspended`, `regional_ingress_required`	Model not allowed for the key or project, organisation suspended, or a request for an ES project that did not enter through the Spain regional endpoint
`429`	`rate_limit_exceeded`, `too_many_concurrent_requests`	Request or concurrency limit; includes `retry-after`
`503`	`no_route`	No allowed route for the project’s tier. Not transient: retrying does not help
`502 / 504`	`upstream_error`, `upstream_timeout`	Upstream failure after all allowed fallbacks

Errors come back in the caller’s dialect: OpenAI format on /v1/chat/… and /v1/embeddings, with the code in error.code; Anthropic format on /v1/messages, with the type in error.type.

 /v1/chat/completions · 503  
 {
  "error": {
    "message": "No route available.",
    "type": "service_unavailable",
    "code": "no_route",
    "param": null
  }
} 

 /v1/messages · 401  
 {
  "type": "error",
  "error": {
    "type": "authentication_error",
    "message": "Invalid API key."
  }
} 

We are looking for three to five organisations for the beta.

Spanish public administrations and regulated organisations that want to use AI models with data residency under control. We work with each organisation on ENS categorisation, the DPA and the integration.

Request beta access Sign up and try it with test keys

Quickstart. From zero to your first request.

Create an account

Create a project

Generate a key