Pay-2-Go (P2G) API

The Pay-2-Go API is the credit-billed API for enterprise and production integrations. Each request is metered against your wallet balance rather than plan quotas.

Base URL

https://api.inferenceport.ai/v1

Authentication

The P2G router resolves either a Supabase JWT or a dashboard-generated API key. All requests require an Authorization: Bearer header.

curl https://api.inferenceport.ai/v1/me \
  -H "Authorization: Bearer YOUR_TOKEN"

Account and config

GET /v1/config

Returns the public dashboard configuration used by the web console.

{
  "dashboard": {
    "appName": "InferencePort AI",
    "homeUrl": "https://inference.js.org",
    "apiBaseUrl": "https://api.inferenceport.ai",
    "supportEmail": "inferenceportai@gmail.com"
  },
  "supabase": {
    "url": "https://...",
    "publishableKey": "...",
    "resetRedirectUrl": "https://inference.js.org/reset.html"
  },
  "billing": {
    "currency": "USD",
    "packs": []
  },
  "pricing": {
    "textCreditPerMillionTokens": 0.75,
    "imageCreditPerImage": 0.02,
    "videoCreditPerSecond": 0.01,
    "audioCreditPerSecond": 0.01,
    "audioDefaultSeconds": 10
  },
  "models": [],
  "notices": [],
  "limits": {
    "requestsPerMinutePerIdentity": 120,
    "minimumTextCreditsBalance": 10
  },
  "generated_at": "2026-06-13T12:00:00Z"
}

GET /v1/me

Returns the current wallet and usage summary for the authenticated identity.

{
  "identity": {
    "email": "user@example.com",
    "auth_type": "jwt",
    "user_id": "uuid-here"
  },
  "wallet": {
    "balance_credits": "42.5000",
    "total_purchased": "100.0000",
    "total_spent": "57.5000"
  },
  "usage_summary": {
    "text": { "total_tokens": 15000, "total_credits": "0.0113" },
    "image": { "total_images": 5, "total_credits": "0.1000" },
    "video": { "total_seconds": 30, "total_credits": "0.3000" },
    "audio": { "total_seconds": 60, "total_credits": "0.6000" }
  },
  "generated_at": "2026-06-13T12:00:00Z"
}

GET /v1/credits/ledger

Returns the latest credit ledger entries.

curl https://api.inferenceport.ai/v1/credits/ledger?limit=50 \
  -H "Authorization: Bearer YOUR_TOKEN"

Query parameters:

• limit — optional integer (default 50, max 500).

Response:

{
  "entries": [
    {
      "id": "uuid",
      "user_id": "uuid",
      "delta_credits": "-0.0113",
      "entry_type": "usage",
      "usage_kind": "text",
      "source": "v1_chat",
      "source_ref": "text:abc123",
      "unit_label": "tokens",
      "metadata": { "input_tokens": 100, "output_tokens": 50, "total_tokens": 150 },
      "created_at": "2026-06-13T12:00:00Z"
    }
  ]
}

Model listing

GET /v1/models

Returns the same model list as /gen/models in OpenAI-compatible format.

{
  "object": "list",
  "data": [ ... ]
}

API key management

GET /v1/lightning-api-keys

Lists all API keys for the authenticated user.

{
  "items": [
    {
      "id": "uuid",
      "name": "My Production Key",
      "keyPrefix": "ipa_live_AbCdEf",
      "createdAt": "2026-01-01T00:00:00Z",
      "lastUsedAt": "2026-06-12T00:00:00Z",
      "expiresAt": null,
      "revokedAt": null,
      "isRevoked": false,
      "isExpired": false
    }
  ]
}

POST /v1/lightning-api-keys

Creates a new API key.

curl -X POST https://api.inferenceport.ai/v1/lightning-api-keys \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "My Production Key",
    "expiresAt": "2027-01-01T00:00:00Z"
  }'

Request fields:

• name — required string (max 64 chars).

• expiresAt — optional ISO-8601 timestamp.

Response includes the rawKey — save it immediately as it is only shown once:

{
  "apiKey": { ... },
  "rawKey": "ipa_live_AbCdEf123456..."
}

DELETE /v1/lightning-api-keys/{key_id}

Revokes an API key immediately.

curl -X DELETE https://api.inferenceport.ai/v1/lightning-api-keys/uuid-here \
  -H "Authorization: Bearer YOUR_TOKEN"

Chat completions

POST /v1/chat/completions

Credit-metered chat completions. The server checks wallet balance before generating and charges credits based on actual token usage.

curl https://api.inferenceport.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      { "role": "user", "content": "Give me a concise API summary." }
    ],
    "model": "lightning",
    "stream": false
  }'

Request fields:

• messages — required array of message objects.

• model — optional string. Defaults to lightning (the auto-router). Specify an override model ID to route to a specific configured provider.

• stream — optional boolean. When true, the response is streamed as text/event-stream.

Streaming behavior:

• The stream includes a payg-usage event (not a data: event) at the end with final credit charge details:

event: payg-usage
data: {"payg_input_tokens": 50, "payg_output_tokens": 120, "payg_total_tokens": 170, "payg_credits_charged": 0.0001275}

Non-streaming response includes payg fields in the usage object:

{
  "id": "...",
  "object": "chat.completion",
  "choices": [ ... ],
  "usage": {
    "prompt_tokens": 50,
    "completion_tokens": 120,
    "total_tokens": 170,
    "payg_input_tokens": 50,
    "payg_output_tokens": 120,
    "payg_total_tokens": 170,
    "payg_credits_charged": 0.0001275
  }
}

Image generation

POST /v1/images/generations

Credit-metered image generation. Charges per image from your wallet.

curl -X POST https://api.inferenceport.ai/v1/images/generations \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A futuristic cityscape at night",
    "n": 1,
    "mode": "fantasy",
    "image_urls": []
  }'

Request fields:

• prompt — required string.

• n — optional integer (1–4). Number of images to generate. Default 1.

• mode — optional string. fantasy or realistic.

• image_urls — optional array of up to 2 URLs or base64 data URIs for image-to-image conditioning.

Response:

{
  "created": 1718300000,
  "data": [
    {
      "b64_json": "/9j/4AAQ...",
      "mime_type": "image/jpeg"
    }
  ],
  "usage": {
    "payg_credits_charged": 0.02,
    "image_count": 1
  }
}

Synchronous Video generation

POST /v1/videos/generations

Warning

The synchronous video generation endpoint is deprecated. Use the async endpoint for better preformance and relaibility, as synchrnous requests will automatically cutoff generations running longer than 100 seconds. See the Async Video Generation API (OpenAI-Compatible) section for more details.

Credit-metered video generation. Charges per second of video from your wallet.

curl -X POST https://api.inferenceport.ai/v1/videos/generations \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Time-lapse of a flower blooming",
    "ratio": "3:2",
    "mode": "normal",
    "duration": 5,
    "image_urls": []
  }'

Request fields:

• prompt — required string.

• ratio — optional string. 3:2 (default), 2:3, or 1:1.

• mode — optional string. normal (default) or fun.

• duration — optional integer. Seconds, clamped to 1–10. Default 5.

• image_urls — optional array of up to 2 URLs or base64 data URIs.

Response is raw video/mp4 bytes with credit charge headers:

Header	Value
X-Payg-Credits-Charged	Decimal credits charged for this request (e.g. 0.05)
X-Payg-Units	Duration in seconds (e.g. 5)
X-Payg-Unit-Label	Always seconds

Audio generation

POST /v1/audio/generations

Credit-metered audio/sound-effect generation. Charges per second of audio from your wallet.

curl -X POST https://api.inferenceport.ai/v1/audio/generations \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Gentle rain and thunder",
    "duration_seconds": 10
  }'

Request fields:

• prompt — required string.

• duration_seconds — optional integer. Clamped to 1–90. Default depends on server configuration (typically 10).

Response is raw audio/mpeg bytes with credit charge headers:

Header	Value
X-Payg-Credits-Charged	Decimal credits charged for this request (e.g. 0.10)
X-Payg-Units	Duration in seconds (e.g. 10)
X-Payg-Unit-Label	Always seconds

Embeddings

POST /v1/embeddings is an OpenAI-compatible text embedding endpoint billed per input token (models from 0.01 to 0.20 credits per million tokens). It accepts model, input, encoding_format and dimensions and returns the credits charged in usage.payg_credits_charged and the X-Payg-* headers. See Embeddings API for the full reference and the model list.

Realtime

WS /v1/realtime?model=... opens a live voice or transcription session using the OpenAI Realtime protocol. Token models (gpt-realtime-2.1, gpt-realtime-2.1-mini) are billed after every response; transcription models (gpt-live-transcribe, scribe-v2-realtime, mai-transcribe-2-streaming) are billed per second of streamed audio. Sessions need at least 0.25 credits to start and close with 4402 when the balance runs out. See Realtime API.

3D generation

3D generation uses an asynchronous job model by default. Submitting a request returns a job_id immediately (HTTP 202); you then poll GET /v1/3d/jobs/{job_id} until the job reaches completed or failed status.

Credits are charged at submission time and refunded automatically if the job fails.

Note

A legacy synchronous mode is available via ?sync=true for existing integrations. It blocks until completion and returns the same response shape as before. It is not recommended for production use because 3D generation can take several minutes and may cause HTTP timeouts on slow connections.

POST /v1/3d/generations

Submit a 3D generation job. Returns 202 Accepted in async mode (default) or 200 OK in sync mode.

# Async (recommended)
curl -X POST https://api.inferenceport.ai/v1/3d/generations \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tripoSR",
    "image_urls": ["https://example.com/object.jpg"]
  }'

# Legacy sync
curl -X POST "https://api.inferenceport.ai/v1/3d/generations?sync=true" \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tripoSR",
    "prompt": "A wooden chair",
    "image_urls": ["https://example.com/chair.jpg"]
  }'

Query parameters:

• sync — optional boolean (default false). Set true to use legacy blocking behaviour.

Request body fields:

• model — required string. "asset-harvester", "tripoSR", sf3d or trellis2.

• image_urls — required array. At least one URL or base64 data URI. The first element is used as the input image.

• prompt — required string when model is "asset-harvester". Not used by TripoSR.

• resolution - Resolution for generation with Trellis 2 - “low”, “medium”, or “high”. Default is “low”. (Optional, only applicable for Trellis 2 model.)

Async response (HTTP 202):

{
  "job_id": "a3f1c2e4b5d6...",
  "status": "pending",
  "created_at": 1718300000.0,
  "poll_url": "/v1/3d/jobs/a3f1c2e4b5d6...",
  "usage": {
    "payg_credits_charged": 0.07, // Varies by model and resolution
    "model_count": 1
  }
}

Sync response (HTTP 200, ?sync=true only):

{
  "created": 1718300000,
  "data": [ { "model_glb_b64_bytes": "<BASE64>" } ],
  "usage": {
    "payg_credits_charged": 0.07, // Varies by model and resolution
    "model_count": 1
  }
}

Pricing

Provider	Model	Price (per Gen)	ID
Nvidia	Asset Harvester	$0.25	asset-harvester
Stability AI	TripoSR	$0.07	tripoSR
Stability AI	SF3D	$0.07	sf3d
Microsoft	Trellis 2 Low Res	$0.24	trellis2
Microsoft	Trellis 2 Medium Res	$0.29	trellis2
Microsoft	Trellis 2 High Res	$0.35	trellis2

Note

When using Trellis 2, to set the resolution, pass the resolution parameter in the request body with one of the following values: “low”, “medium”, or “high”. If not specified, the default resolution is “low”.

GET /v1/3d/jobs/{job_id}

Poll the status of an async 3D job. Returns HTTP 202 while the job is still running and HTTP 200 once it has completed or failed.

curl https://api.inferenceport.ai/v1/3d/jobs/a3f1c2e4b5d6... \
  -H "Authorization: Bearer YOUR_TOKEN"

Possible status values:

Status	Meaning
pending	Job queued, not yet started.
processing	Generation is underway.
completed	Generation finished. data field contains the model output.
failed	Generation failed. error field contains a human-readable reason. Credits have been refunded automatically.

Response while pending or processing (HTTP 202):

{
  "job_id": "a3f1c2e4b5d6...",
  "status": "processing",
  "model": "tripoSR",
  "created_at": 1718300000.0,
  "completed_at": null,
  "usage": { "payg_credits_charged": 0.07, "model_count": 1 } // Price varies by model and resolution
}

Response on completion (HTTP 200):

{
  "job_id": "a3f1c2e4b5d6...",
  "status": "completed",
  "model": "tripoSR",
  "created_at": 1718300000.0,
  "completed_at": 1718300312.5,
  "usage": { "payg_credits_charged": 0.07, "model_count": 1 }, // Price varies by model and resolution
  "data": [
    { "model_glb_b64_bytes": "<BASE64_ENCODED_GLB>" }
  ]
}

Response on failure (HTTP 200):

{
  "job_id": "a3f1c2e4b5d6...",
  "status": "failed",
  "model": "tripoSR",
  "created_at": 1718300000.0,
  "completed_at": 1718300045.1,
  "usage": { "payg_credits_charged": 0.07, "model_count": 1 }, // Price varies by model and resolution
  "error": "TripoSR job abc123 failed: out of memory"
}

Recommended polling interval: 4–10 seconds. Most jobs complete within 1–5 minutes depending on model and server load.

Full async workflow example (Python):

import base64, time, httpx

TOKEN = "YOUR_TOKEN"
BASE  = "https://api.inferenceport.ai/v1"
HEADERS = {"Authorization": f"Bearer {TOKEN}"}

# 1. Submit
r = httpx.post(f"{BASE}/3d/generations", headers=HEADERS, json={
    "model": "tripoSR",
    "image_urls": ["https://example.com/object.jpg"],
})
r.raise_for_status()
job_id = r.json()["job_id"]

# 2. Poll
while True:
    r = httpx.get(f"{BASE}/3d/jobs/{job_id}", headers=HEADERS)
    r.raise_for_status()
    body = r.json()
    if body["status"] == "completed":
        glb_bytes = base64.b64decode(body["data"][0]["model_glb_b64_bytes"])
        break
    if body["status"] == "failed":
        raise RuntimeError(body["error"])
    time.sleep(5)

Output fields in data[]:

• model_glb_b64_bytes — Base64-encoded GLB bytes. Present when model is "tripoSR".

• model_ply_b64_bytes — Base64-encoded PLY (Gaussian Splat) bytes. Present when model is "asset-harvester".

• orbit_video_b64_bytes — Base64-encoded MP4 showing a 360° orbit. Present when model is "asset-harvester".

Note

All _b64_bytes fields contain Base64-encoded raw bytes. Decode them with base64.b64decode() (Python) or Buffer.from(str, 'base64') (Node.js) to obtain the file content.

Note

The 3D generation endpoint is currently in early access and may be subject to change. Contact support if you’re interested in using or providing feedback on this feature.

Stripe

GET /v1/stripe/reconcile/{session_id}

Looks up a Stripe checkout session and returns reconciliation details, including payment method info and receipt URL. Used by the console confirm page.

Pricing

P2G pricing is credit-based. The rates vary by model, resolution, length, and generation type.

The console at https://console.inferenceport.ai always shows the live dashboard values.

Embeddings are billed per million input tokens, realtime conversations per million text/audio/image tokens, and realtime transcription per second of streamed audio; each model’s rates are in GET /v1/embeddings/models and GET /v1/realtime/models (and in the /v1/models catalog).

A minimum balance of 0.30 credits is required before text generation is allowed.

Rate limits

P2G requests are rate-limited per identity at 120 requests per minute by default. Exceeding the limit returns HTTP 429.

Error codes

In addition to the common status codes, the P2G API returns:

• 202 — Job accepted (async 3D generation). Poll /v1/3d/jobs/{job_id} for results.

• 402 — Insufficient credits. The wallet balance is below the minimum required for the operation. Check /v1/me for your current balance and purchase more credits from the console.

• 402 (text) — At least the minimum balance (typically 10 credits) is required for text generation.

• 402 (realtime) — A realtime session needs at least 0.25 credits to start; an open session closes with WebSocket code 4402 when the balance is used up (Realtime API).

• 404 — Job not found (GET /v1/3d/jobs/{job_id}), or unknown embedding / realtime model.

• 504 — Sync 3D request timed out. Use async mode instead.

Decision models

POST /v1/systemone evaluates typed questions with configurable decision providers, including Microsoft Decision 1 on Foundry East US. Discover models at GET /v1/decision/models or in the unified model list. See Decision models / System One API for request examples, provider setup, pricing, console and CLI support.