# WeaveAPI

> An OpenAI-wire-compatible inference gateway. One base URL and one bearer key give access to model
> routes from several upstream families; keys, prepaid balance, usage logs and top-ups are managed in a
> hosted console. Keep your existing OpenAI SDK and change the base URL.

This file was GENERATED by API Evangelist from WeaveAPI's public documentation and from live
unauthenticated probes of its gateway on 2026-08-11. WeaveAPI does not publish an llms.txt of its own
(https://weaveapi.dev/llms.txt returns 404), and publishes no OpenAPI or other machine-readable
contract. Everything below is either quoted from the provider's own pages or recorded from an observed
HTTP response. Nothing here is a WeaveAPI commitment.

## Base URLs

- [OpenAI-compatible base](https://api.weaveapi.dev/v1): use for chat completions, responses, models,
  and every OpenAI-style SDK integration.
- [Anthropic-compatible base](https://api.weaveapi.dev): use for the Claude-style messages route. Do
  NOT append /v1 — Claude-compatible clients build the messages path from the host themselves.

## Authentication

Send a bearer API key created in the console on every request:

```
Authorization: Bearer YOUR_WEAVEAPI_KEY
```

One key grants the account's whole API surface. There are no scopes, no per-key permissions, no test
key mode, and no documented key expiry. A missing key and an invalid key both return the same 401.

## Quick start

```js
import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.WEAVEAPI_KEY,
  baseURL: "https://api.weaveapi.dev/v1",
});
const result = await client.chat.completions.create({
  model: "MODEL_ID_FROM_CONSOLE",
  messages: [{ role: "user", content: "Hello from WeaveAPI." }],
});
```

Model ids must come from the console Model Marketplace or an authenticated `GET /v1/models`. The docs
are explicit that the console — not the website copy — is the source of truth for what is callable.

## Documented endpoints

- `POST /v1/chat/completions` — the broadest compatibility path, recommended for most integrations.
- `POST /v1/responses` — responses-style requests. Confirm the chosen route supports it first; the docs
  warn not to assume every model route does.
- `GET /v1/models` — authenticated catalog of the routes available to your account.

## Undocumented but live endpoints

These routes are registered on the gateway and return an auth challenge rather than a 404, but appear
in no WeaveAPI documentation. Treat them as unsupported until the provider documents them, and verify a
backing model exists before depending on one.

`POST /v1/completions`, `POST /v1/embeddings`, `POST /v1/images/generations`, `POST /v1/audio/speech`,
`POST /v1/audio/transcriptions`, `POST /v1/moderations`, `POST /v1/messages`, `POST /v1/rerank`,
`GET|POST /v1/files`, `GET /v1/realtime`, `GET /v1/dashboard/billing/usage`,
`GET /v1/dashboard/billing/subscription`.

Confirmed absent: `/v1/batches`, `/v1/messages/count_tokens`.

## Models

The public, unauthenticated catalog at `https://api.weaveapi.dev/api/pricing` lists 19 callable routes:

- OpenAI (4): gpt-5.6-luna, gpt-5.6-terra, gpt-5.6-sol, gpt-5.5
- Alibaba Cloud / Qwen (7): qwen3.6-flash, qwen3-coder-next, qwen3.5-plus, qwen3.6-plus, qwen3.7-plus,
  qwen3-coder-plus, qwen3-max-2026-01-23
- Zhipu AI / GLM (4): glm-4.7, glm-5, glm-5.1, glm-5.2
- Moonshot / Kimi (3): kimi-k2.5, kimi-k2.6, kimi-k3
- MiniMax (1): MiniMax-M2.5

WeaveAPI's marketing names seven model families including Anthropic and DeepSeek. The public catalog
contains zero routes from either, though both are registered as vendors in the gateway. Account-specific
access may differ — check the authenticated catalog before planning around a family.

## Pricing

Prepaid credits only; no subscription tiers. Two published paths: self-serve "Pay as you go" and an
"Account review" contact-support motion for larger accounts. New accounts get a $1 signup testing
credit. Per-model rates are published as ratio multipliers (`model_ratio` for input, `completion_ratio`
on top for output, `cache_ratio` for cached input) rather than as dollar amounts; no dollar rate is
published on any public page. Payment is taken via Stripe.

## Errors

Errors use the OpenAI envelope, not RFC 9457:

```json
{"error": {"message": "...", "type": "...", "param": "", "code": ""}}
```

`error.code` is empty on every observed error, so branching requires matching on `message`. Every
response carries an `x-oneapi-request-id` header, which is also embedded in the error message text —
include it when contacting support. Console routes under `/api/*` use a different, undocumented
envelope: `{"message": "...", "success": false}`.

## Operational limits

No numeric rate limits are published, and no `RateLimit-*`, `X-RateLimit-*` or `Retry-After` headers
appear on any observed response. The real ceiling is the prepaid balance. There is no status page, no
changelog, no deprecation policy and no SLA; the terms reserve the right to change model availability,
route names and prices without notice.

## Docs

- [Documentation](https://weaveapi.dev/docs/)
- [Model Marketplace](https://weaveapi.dev/models/)
- [Pricing](https://weaveapi.dev/pricing/)
- [About](https://weaveapi.dev/about/)
- [Support](https://weaveapi.dev/support/)

## Agent tool setup guides

- [Claude Code](https://weaveapi.dev/docs/claude-code-cc-switch/) — Anthropic-compatible, base URL without /v1
- [Codex](https://weaveapi.dev/docs/codex-cc-switch/)
- [OpenCode](https://weaveapi.dev/docs/opencode-cc-switch/)
- [OpenClaw](https://weaveapi.dev/docs/openclaw-cc-switch/)
- [Claude Desktop](https://weaveapi.dev/docs/claude-desktop-cc-switch/)
- [Hermes](https://weaveapi.dev/docs/hermes-cc-switch/)

WeaveAPI hosts no MCP server and publishes no agent skills. The MCP references in the Hermes guide
concern Hermes' own configuration files, not a WeaveAPI endpoint.

## Policies

- [Terms of Service](https://weaveapi.dev/terms/)
- [Privacy Policy](https://weaveapi.dev/privacy/)
- [Abuse Policy](https://weaveapi.dev/abuse/)
- [Refund Policy](https://weaveapi.dev/refund/)

Prompts and responses may be routed to upstream model providers to fulfil requests. WeaveAPI states it
does not sell prompt content, and does not enumerate which upstream handles which route.

## Console

- [Sign up](https://console.weaveapi.dev/register)
- [Sign in](https://console.weaveapi.dev/login)

Console login supports password, GitHub OAuth, Google OIDC and passkeys. API keys themselves support
none of that — they are static bearer strings.

## Contact

support@weaveapi.dev — account, billing, API access and model request issues. When reporting a failed
request include the request time, model id, endpoint path, request id and the full error message. Never
send an API key.