OpenAI-compatible API

Point the official OpenAI SDKs (and most OpenAI-compatible clients) at Hypervize:

base_url = https://hypervize.tech/api
api_key  = hvz_live_…

No /v1 segment is required. Paths live next to our existing routes under /api.

Supported endpoints

Pinned keys (/api/b/…)

Same surface under /api/b/ for keys with pinned_model / pinned_tools (OpenClaw-style stable paths):

/api/b/models, /api/b/chat/completions, /api/b/completions, /api/b/embeddings, /api/b/responses, /api/b/responses/{id}.

SDK examples

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://hypervize.tech/api",
    api_key="hvz_live_…",
)

print(client.models.list())

chat = client.chat.completions.create(
    model="claude-sonnet-5",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)

emb = client.embeddings.create(
    model="cohere.embed-v4",  # or catalog display / value
    input="hello",
)

resp = client.responses.create(
    model="claude-sonnet-5",
    input="Summarize scale-to-zero in one sentence.",
)
print(resp.output_text)  # if SDK exposes it; else read resp.output

Node

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://hypervize.tech/api",
  apiKey: process.env.HYPERVIZE_API_KEY,
});

const models = await client.models.list();
const completion = await client.chat.completions.create({
  model: "claude-sonnet-5",
  messages: [{ role: "user", content: "Hello" }],
});

Models + pricing

GET /api/models returns OpenAI list objects plus Hypervize fields:

• value, provider, chat_eligible, max_tokens, context_window
• pricing: { currency, input_per_million, output_per_million, per_image, display } from the live prices table (null if unset)

Primary id is the display name (e.g. claude-sonnet-5). Request bodies accept display or full catalog value.

Responses API notes

• Tools:
  • Plain client tools (no webhook / not platform): mapped to Responses function_call output items (stream and non-stream). Plain-tool requests do not attach platform tools unless hybrid is enabled (same as chat completions).
  • Platform / webhook tools: managed path (final text and/or tool execution server-side), same as chat completions Mode B.
  • Both together: see Using plain tools with platform tools.
• Never rely on empty status: "completed" with blank text as a tool result — that was a bug; you should see function_call items or a real error.
• previous_response_id: continues from a prior response (~7 day retention). Works for non-stream and stream (stream is stored after response.completed when the client drains the body).
• store: false: do not persist (cannot chain later).
• Chain context is bounded to limit growth on long multi-turn sessions.
• Streaming event subset: response.created, response.output_text.delta, response.completed (+ [DONE]). function_call items appear on response.completed.output.

Completions (legacy)

POST /api/completions maps prompt → a single user chat message, returns choices[].text.
Platform tools are not injected on this path. Prefer chat or responses for agent/tool use.

Errors (dual shape)

Inference errors use OpenAI-style objects and a top-level message for older clients:

{
  "error": {
    "message": "Insufficient prepaid balance. Please add funds to continue.",
    "type": "insufficient_quota",
    "code": "insufficient_balance",
    "param": null
  },
  "message": "Insufficient prepaid balance. Please add funds to continue.",
  "current_balance_usd": 0
}

Prefer error.message or top-level message. Do not assume error is a string.

Billing

Same prepaid ledger as elastic chat: free queries, balance gates, usage queue, platform tools metering. Dedicated weekly post-pay is separate (/api/d/…).

See Billing & Payments.