Quickstart

• Create an account, then open the console and mint a key for your organization.
• Every key belongs to one organization. A key cannot read another organization’s stored responses, files, or usage.
• Store it in the environment, never in client-side code. The console can revoke any key immediately; revoked keys stop authenticating.
• Platforms that provision keys for tenants use a separate machine path — see platform keys.

export SYLPHX_API_KEY="sk-sx-…"  # paste the key from the console

curl https://api.models.sylphx.ai/v1/responses \
  -H "Authorization: Bearer $SYLPHX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.5",
    "input": "Explain streaming in two sentences."
  }'

Only two things are Sylphx-specific: the base URL and the model id. Swap model for any id in the catalog and the rest of your request is unchanged.

{
  "id": "resp_9f2c41d0a8",
  "object": "response",
  "model": "openai/gpt-5.5",
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "Streaming sends each token as the model writes it…",
          "annotations": []
        }
      ]
    }
  ],
  "usage": {
    "input_tokens": 18,
    "output_tokens": 96,
    "total_tokens": 114
  }
}

• id identifies this response. While it is stored you can retrieve it with GET /responses/{id}, list its input items, or delete it.
• model is exactly the model id you sent — never an internal alias. That is what makes the price and the transcript reproducible.
• output carries the assistant message, and any function calls or hosted tool items the turn produced.
• usage reports the tokens this request was metered for. Multiply by the catalog’s per-million prices to reproduce the cost.

curl -N https://api.models.sylphx.ai/v1/responses \
  -H "Authorization: Bearer $SYLPHX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5.5",
    "input": "Count to five.",
    "stream": true
  }'

event: response.created
data: {"type":"response.created","response":{…}}

event: response.output_text.delta
data: {"type":"response.output_text.delta","delta":"One"}

event: response.completed
data: {"type":"response.completed","response":{…},"usage":{…}}

Streaming rules — one terminal, ordered events, and tool calls that do not end the stream — are on the Responses page.

curl https://api.models.sylphx.ai/v1/responses \
  -H "Authorization: Bearer $SYLPHX_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: 8f8d57c8-8868-4d90-90cb-55e892cb80a5" \
  -d '{
    "model": "openai/gpt-5.5",
    "input": "Summarise this incident in three bullets."
  }'

• Identical retry after completion → 200 with the original response and header idempotency-replayed: true.
• Same key, different body → 422 idempotency_key_reuse. Use a new key for a new request.
• Same key while the first attempt is still running → 409 idempotency_in_progress. Wait, then retry the same bytes.

Full retry contract, including which errors are retryable: errors and retries.

• 401 invalid_api_key — the bearer is missing, malformed, or revoked. Check that the value starts with sk-sx- and has no quotes or whitespace.
• 404 model_not_found — that model id is not in the catalog. Copy an id from the catalog rather than typing one.
• 429 rate_limit_exceeded — the key’s request envelope is spent. Read retry_after_seconds (and the Retry-After header) before retrying.

Every error shares one envelope, and each code names the next action. See errors and retries.