NeuralSpace API

Language: Русский · English · العربية · 简体中文 · ko · ja · Deutsch · Français · Español · Italiano · Svenska · עברית · ייִדיש · हिन्दी · Nederlands · Português · Română · Polski · Čeština · Ελληνικά · Slovenčina · Magyar · Bahasa Indonesia · Bahasa Melayu · Dansk · Suomi · Norsk · Hrvatski · sr

Documentation for developers and AI agents. This page is available without authentication and is intended, among other things, to be read automatically.

Last updated: September 7, 2026

Base URL: https://neuralspace.pro. You pay with account balance tokens (1 token = 1 ₽), no subscriptions or foreign cards required. The text and chat endpoints are compatible with the Anthropic and OpenAI SDKs — just change base_url and the key.

Authentication

Create an API key in your account: https://neuralspace.pro/api-keys. The key looks like nsk-... and is shown only once. Pass it in the header:

x-api-key: nsk-your_key

The Authorization: Bearer nsk-... header is also accepted (for OpenAI-compatible clients).

• Rate limit: 60 requests per minute per key.
• Minimum balance to access the text models: 50 tokens.
• Streaming (stream: true) is supported on /v1/messages and /v1/chat/completions: the response arrives via Server-Sent Events in the format of the corresponding SDK (Anthropic / OpenAI). With stream: false you get a regular JSON response.

MCP server — connect NeuralSpace to your AI agent

Everything this API can do (image/video/music generation, chatting with LLMs, speech synthesis and recognition, balance, prices) is also available as an MCP server (Model Context Protocol) — you can connect it to Claude Code, Claude Desktop, Cursor and any other MCP client, and your AI agent will generate content itself on behalf of your account.

Endpoint (Streamable HTTP, sessionless): https://neuralspace.pro/mcp. Authorization — the same nsk-... key in the Authorization: Bearer or x-api-key header.

# Claude Code
claude mcp add --transport http neuralspace https://neuralspace.pro/mcp \
  --header "Authorization: Bearer nsk-your_key"

// Cursor / Claude Desktop and other streamable-http clients
{
  "mcpServers": {
    "neuralspace": {
      "type": "http",
      "url": "https://neuralspace.pro/mcp",
      "headers": { "Authorization": "Bearer nsk-your_key" }
    }
  }
}

// Clients that only speak stdio — via the mcp-remote bridge
{
  "mcpServers": {
    "neuralspace": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://neuralspace.pro/mcp",
               "--header", "Authorization: Bearer nsk-your_key"]
    }
  }
}

MCP server tools: list_models (models and live prices), get_balance (balance), chat (LLM), generate_image, generate_video + check_video, generate_music + check_music, text_to_speech, transcribe_audio. Billing and refunds work exactly as for direct API calls; successful generations appear as cards in your account. For synchronous image generation (up to ~2 minutes) increase the MCP tool timeout in your client if it is lower.

GET /v1/balance and GET /v1/models — service endpoints

GET /v1/balance — the current token balance of the key owner: { "balance": N, "currency": "tokens" }. GET /v1/models — a machine-readable list of all models with live prices (the JSON version of this page). Both require a key.

POST /v1/messages — text models (Anthropic format)

Compatible with the Anthropic Messages API. Models: claude-haiku-4-5, claude-haiku-5-5, claude-sonnet-5, claude-sonnet-5-5, claude-opus-5, claude-opus-5-5, claude-fable-5, claude-fable-5-1.

Body fields: model, messages (array of {role, content}, role — user/assistant), max_tokens (1—32000), optional system.

curl https://neuralspace.pro/v1/messages \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-opus-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Via the Anthropic SDK (Python): Anthropic(base_url="https://neuralspace.pro", api_key="nsk-...").

POST /v1/chat/completions — chat models (OpenAI format)

Compatible with the OpenAI Chat Completions API. Models: gpt-5-6, gpt-5-6-balanced, gpt-5-6-fast, gpt-6-astra, gpt-6-sol, gpt-6-luna, gpt-6-1-sol, gpt-6-1-sol-pro, gpt-6-1-terra, gpt-6-1-terra-pro, gpt-5-mini, gpt-5-4-mini, grok-4-3, grok-4-5, grok-4-6, grok-4-7, gemini-3-8-flash, gemini-4-argon, deepseek-v4.1-flash, deepseek-v4.1-flash-1017, deepseek-v4-pro, glm-5.3-flash, glm-5.3, kimi-k2.6, kimi-k3, kimi-k3.1-pro.

Body fields: model, messages (role — system/user/assistant), max_tokens or max_completion_tokens, optional reasoning_effort (for GPT-5.6, GPT-6 Astra, and Grok 4.5).

curl https://neuralspace.pro/v1/chat/completions \
  -H "Authorization: Bearer nsk-your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-5-6-balanced",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Via the OpenAI SDK: OpenAI(base_url="https://neuralspace.pro/v1", api_key="nsk-...").

Realtime voice

Grok Think Fast is available over WebSocket for native speech-to-speech. Sessions are limited to 120 minutes. Billing uses the actual duration of input and output audio; the exact amount is in the price table above.

const ws = new WebSocket(
  "wss://neuralspace.pro/v1/realtime?model=grok-voice-think-fast-2.0&lang=ja",
  { headers: { Authorization: "Bearer nsk-your_key" } }
);

POST /v1/images/generations — image generation

Synchronous: the response contains a ready URL. Tokens are charged only on success. Models: nano-banana, nano-banana-pro, nano-banana-2, nano-banana-2-lite, nano-banana-2-1, flux-2-pro, gpt-image-15, gpt-image-2, gpt-image-2-pro, gpt-image-25, seedream-45, seedream-5-lite, seedream-5-flash, seedream-5-pro, grok-image, grok-imagine-image-2, ns-image, z-image, ideogram, midjourney.

Body fields: model, prompt (up to 5000 characters), optional aspect_ratio (gpt-image-2-pro: 1:1/3:2/2:3; gpt-image-25: auto/1:1/3:2/2:3/16:9/9:16/4:3/3:4/21:9, with 27:16/16:27/9:8/8:9 at 1K only), resolution (1K/2K/4K), variant (flare/sunburst — gpt-image-25), background (transparent/opaque/auto — gpt-image-25), output_format (jpeg/png — nano-banana-pro; jpeg/png/webp — gpt-image-2-pro), quality (medium/high — gpt-image-15; low/medium/high — gpt-image-2-pro), mode (relaxed/fast/turbo — midjourney), image_urls (array — enables edit / image-to-image mode).

curl https://neuralspace.pro/v1/images/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "nano-banana-pro",
    "prompt": "A cat in space, digital illustration",
    "resolution": "2K"
  }'

Response: { "data": [{ "url": "https://..." }], "usage": { "tokens": N } }.

Input images: URL or Base64

Each item of image_urls is either a URL or Base64: a data URL like data:image/png;base64,... or plain base64 without a prefix. The image is decoded, verified and stored on the server; the models then receive it as a regular URL — the response and billing are identical to passing a link. Formats: PNG/JPEG/WebP; up to 10 MB per image and up to 72 MB of media per request. Invalid format, a declared type that does not match the file content, or exceeding the limit — a 400 error, no tokens are charged.

# 2. image_urls as a URL OR base64 (data:image/png;base64,...)
curl https://neuralspace.pro/v1/images/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "nano-banana-2-1",
    "prompt": "make the cat wear a space helmet",
    "image_urls": ["data:image/png;base64,iVBORw0KGgoAAAANSUhEUg..."]
  }'

POST /v1/videos/generations — video generation (asynchronous)

Video takes minutes to generate, so it is asynchronous: POST creates a task and charges tokens, then poll GET /v1/videos/generations/<id>. On error/timeout the tokens are refunded. Models: seedance-2.0, seedance-2.5, veo-3.1, veo-3.1-fast, veo-3.1-lite, sora-2, sora-2-pro, kling-o3, kling-3.0, kling-3-0-turbo, kling-4-0, grok-1.5, grok-imagine, wan-2.6, gemini-omni-flash-1-1, gemini-omni-1, vidu-q4.

Body fields: model, prompt, optional duration, resolution, speed (seedance), mode (kling/grok), quality (sora-2-pro), audio (kling), aspect_ratio, image_urls, multi_shots (storyboard — kling-3.0 and wan-2.6). Grok Imagine Video 1.5: model: "grok-1.5", 1–15 seconds, 480p/720p/1080p; enable_audio (true/false), first_frame_url, last_frame_url, loop_url (same start and end image), keyframes: [{"url":"https://example.com/frame.jpg","timestamp_s":3}]. Up to 4 mid-video frames at distinct whole seconds greater than 0 and less than duration. A prompt is optional when at least one frame is supplied. Images accept URLs or Base64.

Additionally for kling-3.0:

kling-4-0: duration is an integer from 3 to 30, resolution is 720p/1080p, and aspect_ratio is auto/16:9/9:16/1:1/21:9. reference mode accepts up to 10 image_urls, 5 video_urls, 7 elements ({name, image_urls or video_url, tag, description, voice_url}), and 3 voices ({name, audio_url}), with at most 15 image/video references total. Media in these fields can be URLs or Base64. keep_original_sound needs an input video. keyframe mode accepts first_frame_url, last_frame_url, and up to 8 keyframe_urls; modes cannot be combined.

• multi_shots (true) + shots — a storyboard of 2—5 shots; each shot is a prompt string or an object {prompt, duration} (with multi_shots at most one image is allowed in image_urls);
• elements — an array (up to 4) of consistent characters/objects; reference an element in the prompt as @element_<name>. Each element: { "name": "...", "description": "...", "image_urls": [2—4 photos] } or, instead of photos, "video_url": "..." (1 video).

# 1. create the task
curl https://neuralspace.pro/v1/videos/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{"model": "seedance-2.0", "prompt": "Sunset over the ocean", "duration": 5}'
# response: { "id": "task_...", "status": "pending", "poll_url": "..." }

# 2. poll the status until status != completed/failed
curl https://neuralspace.pro/v1/videos/generations/task_... \
  -H "x-api-key: nsk-your_key"
# when completed: { "status": "completed", "data": [{ "url": "https://..." }] }

Input video: URL or Base64

Input videos can also be sent as Base64: items of video_urls/reference_video_urls, video_list[].url, elements[].video_url — either a URL or a data URL data:video/mp4;base64,..., or plain base64. Formats: MP4/WebM/MOV; up to 64 MB per video (within the 72 MB total per-request limit). The same checks apply: file signature, declared type that does not match the content — a 400 error, no tokens are charged.

# reference video as a URL OR base64 (data:video/mp4;base64,...)
curl https://neuralspace.pro/v1/videos/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{"model": "seedance-2.5", "prompt": "continue this clip", "duration": 5,
       "video_urls": ["data:video/mp4;base64,AAAAIGZ0eXBpc29t..."]}'

POST /v1/music/generations — music generation (asynchronous)

Music (Suno) takes minutes to generate, so it is asynchronous: POST creates a task and charges tokens, then poll GET /v1/music/generations/<id>. On error/timeout the tokens are refunded. One generation usually returns 2 tracks. Models: V6, V6_WILD, V6_MINI, V5_5, V5, V4_5PLUS, V4_5, V4.

Body fields: model, prompt (description of the music; in custom_mode with vocals — the song lyrics), optional instrumental (no vocals), custom_mode (requires style and title), style, title, negative_tags, vocal_gender (m/f), duration — track length in seconds, from 10 to 360 (available for the V5_5, V6, V6_MINI and V6_WILD models together with custom_mode: true; without it the model chooses the length).

# 1. create the task
curl https://neuralspace.pro/v1/music/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{"model": "V5", "prompt": "Energetic synthwave about a night city"}'
# response: { "id": "task_...", "status": "pending", "poll_url": "..." }

# 2. poll the status until status != completed/failed
curl https://neuralspace.pro/v1/music/generations/task_... \
  -H "x-api-key: nsk-your_key"
# when completed: { "status": "completed", "data": [{ "url": "https://...", "title": "...", "duration": 123 }] }

POST /v1/audio/speech — speech synthesis (OpenAI format)

Compatible with the OpenAI Audio Speech API. Synchronous: the response contains a ready audio file URL. Tokens are charged only on success, by number of characters. Models: tts-1, tts-1-hd.

Body fields: model, input (text, up to 4096 characters), optional voice (alloy/echo/fable/onyx/nova/shimmer), response_format (mp3/opus/aac/flac/wav).

curl https://neuralspace.pro/v1/audio/speech \
  -H "Authorization: Bearer nsk-your_key" \
  -H "content-type: application/json" \
  -d '{
    "model": "tts-1",
    "voice": "alloy",
    "input": "Hello! This is NeuralSpace speech synthesis."
  }'

POST /v1/audio/transcriptions — audio transcription (OpenAI format)

Compatible with the OpenAI Audio Transcriptions API (Whisper). The file is sent as multipart/form-data. Synchronous: the response contains the recognized text. Tokens are charged only on success, by audio duration (estimated from file size). Models: whisper-large-v3, whisper-1. File limit — 25 MB.

Form fields: file (audio: mp3/mp4/mpeg/mpga/m4a/wav/webm/ogg/flac), model (whisper-large-v3; the legacy whisper-1 is accepted and run as large-v3), optional language (ISO code, e.g. ru/en), prompt (context hint).

curl https://neuralspace.pro/v1/audio/transcriptions \
  -H "Authorization: Bearer nsk-your_key" \
  -F model="whisper-large-v3" \
  -F file="@audio.mp3"
# response: { "text": "...", "model": "whisper-large-v3", "usage": { "tokens": 0.2 } }

Response: { "text": "recognized text", "model": "whisper-large-v3", "usage": { "tokens": 0.2 } }.

動画アバター

話す動画アバターと関連モード。ディレクトリ: GET /v1/avatars（avatar_id）、GET /v1/avatars/voices?language=en（voice_id）、GET /v1/avatars/languages（翻訳言語）、/v1/avatars/styles、/v1/avatars/templates、/v1/avatars/assets、/v1/avatars/brand-kits。生成は非同期です。type フィールドを付けて POST /v1/avatars/generations を呼び出し、その後 GET /v1/avatars/generations/<id> でポーリングします。トークンは作成時に消費され、失敗時には返金されます。

curl https://neuralspace.pro/v1/avatars/voices?language=en \
  -H "x-api-key: nsk-your_key"

curl https://neuralspace.pro/v1/avatars/generations \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{"type": "talking_avatar", "avatar_id": "avatar_...", "voice_id": "voice_...", "script": "Hello!"}'
# response: { "id": "task_...", "status": "pending", "poll_url": "..." }

POST /v1/characters — create a character / voice assistant

Creates an AI character or voice assistant in your account (available in the “Characters” section). Without image_url a voice assistant (voice_only) with a preset voice is created — fast and cheap. With image_url (a public photo URL) a character with a video avatar is created. Cost — 10 tokens, charged on success.

Fields: name (required), instruction (persona / system instruction), voice (for voice assistant: alloy, echo, shimmer, sage, verse, coral, ash, ballad; default alloy), start_script (greeting), image_url (photo for a video avatar), language (ru/en).

curl https://neuralspace.pro/v1/characters \
  -H "Authorization: Bearer nsk-your_key" \
  -H "Content-Type: application/json" \
  -d '{ "name": "My assistant", "instruction": "A friendly product assistant", "voice": "verse" }'
# response: { "character": { "id": "voice-...", "name": "...", "characterType": "voice_only", "voice": {...} } }

GET /v1/characters — list your characters

Returns { "data": [{ "id", "name", "characterType", "voice", "isVoiceOnly", ... }] }.

Models and prices

Prices are in site tokens (1 token = 1 ₽). Rates are dynamic and may change; this page always reflects the current values.

Prices float: each request is charged its actual cost, so the figures in the tables are a reference, not a fixed rate. Where a model price moves during the day, the range over the last 24 hours is shown below it.

Text models (per 1M tokens)

Chat models (per 1M tokens)

Images (per image)

Video

Music

Speech synthesis (per 1000 characters)

Audio transcription (per 1 minute)

動画アバター

3D models (Tripo)

3D generation and model processing: 18 operations (3D from text/photo/views, splat, images, texture, segmentation, completion, retopology, UV, format conversion, rigging, animations). Operation codes and parameters: GET /v1/tripo/tools; token prices use the live rate. Asynchronous creation: POST /v1/tripo/tasks (tool, prompt, params, image_urls, animations), then poll GET /v1/tripo/tasks/<id>. Results are stored in our storage with permanent links. Tokens are charged on creation and refunded on failure.

curl https://neuralspace.pro/v1/tripo/tools \
  -H "x-api-key: nsk-your_key"

curl https://neuralspace.pro/v1/tripo/tasks \
  -H "x-api-key: nsk-your_key" \
  -H "content-type: application/json" \
  -d '{"tool": "text-to-3d", "prompt": "a wooden chair"}'
# response: { "id": "...", "status": "pending", "poll_url": "/v1/tripo/tasks/..." }

Errors

The text endpoint returns errors in the Anthropic format ({ "type": "error", "error": { "type", "message" } }), chat — in the OpenAI format ({ "error": { "message", "type" } }). Codes: 401 — missing/invalid key, 403 — account suspended, 429 — rate limit exceeded, 400 — invalid request or insufficient tokens.

Support and balance top-up: https://neuralspace.pro/payment. Manage keys: https://neuralspace.pro/api-keys.

Ready-to-copy settings for OpenCode, Claude Code, Cursor, Harness and MCP.