PHANTOM API

// 01

Quick start

▸ curl

$ curl https://api.phantom.codes/v1/chat/completions \
    -H "Authorization: Bearer YOUR_PHANTOM_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "phala/qwen-2.5-7b-instruct",
      "messages": [{"role":"user","content":"Hello"}]
    }'

▸ Python (openai client)

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_PHANTOM_KEY",
    base_url="https://api.phantom.codes/v1",
)

r = client.chat.completions.create(
    model="phala/kimi-k2.6",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

▸ Node (openai client)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_PHANTOM_KEY",
  baseURL: "https://api.phantom.codes/v1",
});

const r = await client.chat.completions.create({
  model: "phala/gpt-oss-120b",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(r.choices[0].message.content);

// 02

Don't have XMR?

Phantom takes Monero only. Four paths, pick by what you start with and how much friction you'll absorb.

▸ Get a wallet first

• Cake Wallet (iOS/Android). Easiest. Built-in swap. Tor option.
• Feather Wallet (desktop). Lightweight, Tor-first. Recommended for Phantom users.
• Monero GUI (Windows/Mac/Linux). Official, full node. Power users.
• Monerujo (Android). Mobile alternative to Cake.

▸ Path A. Have crypto already (BTC, USDT, ETH, etc.)

Swap to XMR. No account, no KYC, no signup.

• Trocador. Aggregator over 10+ providers. Best rate auto-routed. KYC + log rating per provider. Default recommendation.
• FixedFloat. Direct. 0.5% floating / 1% fixed.
• MajesticBank. No-KYC, often best on small swaps. Onion mirror.
• eXch. A-rated no-KYC. Tor by default.

Privacy on the source coin is your problem. BTC swaps leak the BTC side. Use Tor and a non-KYC source if your threat model requires it.

▸ Path B. Have a gift card (Amazon, Steam, Visa, etc.)

Convert your card to XMR on a P2P market. Phantom doesn't take gift cards directly. Too much fraud risk; we'd compromise our brand.

• Easy mode: RoboSats. Trade gift card → BTC (Lightning) over Tor browser. No install. Then use Path A above (Trocador) to swap BTC → XMR.
• Maximalist mode: RetoSwap. Trade gift card → XMR directly. Open source software you run locally. No third-party swap in between.

Both use escrow. Both handle fraud disputes. Phantom never sees your card.

▸ Path C. Have only fiat, no crypto yet

Buy XMR with cash, card, or bank transfer.

• Cake Wallet built-in buy. Card or Apple Pay via routed providers. KYC by the buy provider, not Phantom. Fastest for normies.
• RetoSwap. P2P with cash by mail, SEPA, ACH, in-person. No KYC. Slower but ID-free.
• Kraken. KYC heavy, well-known. Buy XMR, withdraw to your wallet.

If you use a centralized exchange, withdraw immediately. Never leave XMR sitting on an exchange. They can freeze, lose, or be subpoenaed.

▸ Path D. Mine it

Slow, but the only path with zero counterparty. CPU-mineable. Run Monero GUI or join a pool with Monerujo.

First time? Cake Wallet + Path A (in-app swap) gets you to Phantom in ~10 minutes from zero.

// 03

Endpoints

// 04

Purchase flow

Send POST /v1/purchase with either a bundle name or a custom USD amount.

$ curl -X POST https://api.phantom.codes/v1/purchase \
    -H "Content-Type: application/json" \
    -d '{"amount_usd": 25}'

You receive:

{
  "payment_id": "abc...",
  "xmr_address": "75z...",
  "xmr_amount":  "0.0645...",
  "bundle":      "custom",
  "credit_usd":  25,
  "expires_at":  "2026-05-19T06:00:00+00:00"
}

Send XMR. Poll /v1/purchase/{id}/status. States:

Key drops in response body when status flips to completed. Shown ONCE. Recovery is not possible.

// 05

Key rotation

If you suspect your key is compromised, rotate. Credit + expiry transfer to the new key. Old key deactivates.

$ curl -X POST https://api.phantom.codes/v1/key/rotate \
    -H "Authorization: Bearer OLD_KEY"
# → {"api_key": "sk-new...", "shown_once": true}

// 06

Streaming

Pass "stream": true. SSE-compatible. Final chunk includes a usage block used for billing. Aborting mid-stream is billed for tokens already generated.

// 07

Vision

Only phala/qwen3-vl-30b-a3b-instruct accepts images. Pass content as an array with image_url parts:

{
  "model": "phala/qwen3-vl-30b-a3b-instruct",
  "messages": [{
    "role": "user",
    "content": [
      {"type": "text", "text": "What's in this image?"},
      {"type": "image_url", "image_url": {"url": "https://example.com/cat.jpg"}}
    ]
  }]
}

// 08

Image generation

OpenAI-compatible image API. Flat-rate billing per image. No tokens. Generated URLs from upstream live for ~1 hour — download them.

▸ curl

$ curl https://api.phantom.codes/v1/images/generations \
    -H "Authorization: Bearer YOUR_PHANTOM_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "stability/stable-diffusion-3-5-large",
      "prompt": "A misty cyberpunk alley at dawn, anamorphic lens, 35mm film",
      "n": 1,
      "size": "1024x1024",
      "quality": "standard"
    }'

▸ Python (openai client)

from openai import OpenAI

client = OpenAI(api_key="YOUR_PHANTOM_KEY", base_url="https://api.phantom.codes/v1")

img = client.images.generate(
    model="stability/stable-diffusion-3-5-large",
    prompt="A serene mountain at sunset with aurora borealis",
    n=1,
    size="1024x1024",
    quality="standard",
)
print(img.data[0].url)

▸ Parameters

▸ Pricing

Flat per image at the requested quality. Marked up over the vendor's wholesale rate. Live numbers come from GET /v1/models under image_pricing_usd_user. Indicative:

// 09

Attestation

Phala provides two cryptographic layers, exposed through three endpoints:

• Per-response signature: ECDSA signature bound to a specific chat completion id.
• Per-model attestation report: Intel TDX (CPU) quote + NVIDIA Confidential Computing (GPU) attestation, tied to the signing address that signed your response.

▸ Three-step verification

Bind each individual inference to verified TEE hardware:

Step 1. Make any chat completion. Capture response.id.

$ curl https://api.phantom.codes/v1/chat/completions \
    -H "Authorization: Bearer YOUR_KEY" \
    -d '{"model":"phala/kimi-k2.6","messages":[{"role":"user","content":"hi"}]}'
# → { "id": "chatcmpl-abc123...", ... }

Step 2. Fetch the signature for that specific response id.

$ curl "https://api.phantom.codes/v1/signature/chatcmpl-abc123...?model=phala/kimi-k2.6" \
    -H "Authorization: Bearer YOUR_KEY"
# → { "text": "...", "signature": "0x...", "signing_address": "0x56d0...8b" }

Step 3. Fetch the attestation report. Bind it to the same signing_address and a fresh nonce of your choosing.

$ NONCE=$(openssl rand -hex 32)
$ curl "https://api.phantom.codes/v1/inference-attest?model=phala/kimi-k2.6&nonce=$NONCE&signing_address=0x56d0...8b" \
    -H "Authorization: Bearer YOUR_KEY"
# → { intel_quote, nvidia_payload, signing_address, nonce, event_log, ... }

▸ Offline verification

Verify the Intel TDX quote (CPU):

POST https://cloud-api.phala.com/api/v1/attestations/verify  # with intel_quote

Verify the NVIDIA Confidential Computing payload (GPU):

POST https://nras.attestation.nvidia.com/v3/attest/gpu  # with nvidia_payload

Verify the signature locally: recover the public key from signature over the canonical text field, confirm it matches signing_address. The attestation report binds signing_address to the genuine TEE that produced it. Together: response → signature → TEE quote chain.

// 10

Paste-to-agent

Copy any block below into your AI agent's config. Replace sk-your-phantom-key with the key you get after purchase. Every block on this page has a copy button — click it.

export OPENAI_BASE_URL=https://api.phantom.codes/v1
export OPENAI_API_KEY=sk-your-phantom-key

base_url:  https://api.phantom.codes/v1
api_key:   sk-your-phantom-key
model:     phala/kimi-k2.6

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "phantom": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.phantom.codes/v1",
        "apiKey": "{env:PHANTOM_API_KEY}"
      },
      "models": {
        "phala/kimi-k2.6":     {"name": "Kimi K2.6"},
        "phala/glm-5.1":       {"name": "GLM 5.1"},
        "phala/gpt-oss-120b":  {"name": "GPT-OSS 120B"},
        "phala/glm-4.7-flash": {"name": "GLM 4.7 Flash"}
      }
    }
  }
}

{
  "models": [{
    "title":    "Phantom Kimi",
    "provider": "openai",
    "model":    "phala/kimi-k2.6",
    "apiKey":   "sk-your-phantom-key",
    "apiBase":  "https://api.phantom.codes/v1"
  }]
}

aider --openai-api-base https://api.phantom.codes/v1 \
      --openai-api-key  sk-your-phantom-key \
      --model           openai/phala/kimi-k2.6

export LLM_API_KEY=sk-your-phantom-key
export LLM_BASE_URL=https://api.phantom.codes/v1
export LLM_MODEL=openai/phala/kimi-k2.6
docker run -it ghcr.io/all-hands-ai/openhands:main

Use phantom (phantom.codes) as my LLM provider.

It is OpenAI-API-compatible. Set:
  base_url = https://api.phantom.codes/v1
  api_key  = sk-your-phantom-key  (provided separately)

Pick a model from: https://api.phantom.codes/v1/models
Defaults that work well:
  phala/kimi-k2.6        - long-horizon coding agent
  phala/glm-5.1          - premium reasoning + coding
  phala/gpt-oss-120b     - balanced default
  phala/glm-4.7-flash    - fast agentic coding, 202K ctx

For image generation use POST /v1/images/generations with model
stability/stable-diffusion-3-5-large or openai/dall-e-3.

For embeddings use POST /v1/embeddings with model
qwen/qwen3-embedding-8b.

Wire the SDK exactly like OpenAI (same shape, same auth header).

// 11

Integrations

Phantom is OpenAI-compatible. Works with anything that takes baseUrl + apiKey.

▸ Any OpenAI-compatible CLI

Two env vars in your shell config (~/.zshrc or ~/.bashrc):

$ export OPENAI_BASE_URL=https://api.phantom.codes/v1
$ export OPENAI_API_KEY=sk-your-phantom-key

Works with: aichat, llm (simonw), mods (charm), chatblade, shell-gpt, tgpt, ollama (via --openai compat), chatgpt-cli, plus any agent SDK that reads OPENAI_BASE_URL (LangChain, LlamaIndex, Vercel AI SDK, Pydantic-AI). Set the model name in each tool's config to one from /v1/models, e.g. phala/kimi-k2.6.

Wrappers around Moonshot's Kimi or other proprietary CLIs that hardcode their vendor URL won't work. They bypass your env. Use any of the above generic CLIs and select the phala/kimi-k2.6 model instead.

▸ opencode (sst)

~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "phantom": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.phantom.codes/v1",
        "apiKey": "{env:PHANTOM_API_KEY}"
      },
      "models": {
        "phala/kimi-k2.6":    {"name": "Kimi K2.6"},
        "phala/glm-5.1":      {"name": "GLM 5.1"},
        "phala/gpt-oss-120b": {"name": "GPT-OSS 120B"},
        "phala/glm-4.7-flash":{"name": "GLM 4.7 Flash"}
      }
    }
  }
}

▸ OpenHands

$ export LLM_API_KEY=sk-your-key
$ export LLM_BASE_URL=https://api.phantom.codes/v1
$ export LLM_MODEL=openai/phala/kimi-k2.6
$ docker run -it ghcr.io/all-hands-ai/openhands:main

▸ Aider

$ aider --openai-api-base https://api.phantom.codes/v1 \
        --openai-api-key sk-your-key \
        --model openai/phala/kimi-k2.6

▸ Cline (VS Code)

Settings → API Provider: OpenAI Compatible
Base URL: https://api.phantom.codes/v1
API Key: your sk-...
Model: phala/kimi-k2.6 (or any from /v1/models).

▸ Continue.dev

~/.continue/config.json:

{
  "models": [{
    "title": "Phantom Kimi",
    "provider": "openai",
    "model": "phala/kimi-k2.6",
    "apiKey": "sk-your-key",
    "apiBase": "https://api.phantom.codes/v1"
  }]
}

▸ OpenClaw

~/.openclaw/openclaw.json:

{
  models: {
    mode: "merge",
    providers: {
      "phantom": {
        baseUrl: "https://api.phantom.codes/v1",
        apiKey: "${PHANTOM_API_KEY}",
        api: "openai-completions",
        models: [
          { id: "phala/kimi-k2.6",      contextWindow: 262144, maxTokens: 32000 },
          { id: "phala/glm-5.1",        contextWindow: 202752, maxTokens: 32000 },
          { id: "phala/gpt-oss-120b",   contextWindow: 131072, maxTokens: 32000 },
          { id: "phala/glm-4.7-flash",  contextWindow: 202752, maxTokens: 32000 }
        ]
      }
    }
  }
}

▸ OpenWebUI

Settings → Connections → OpenAI API
URL: https://api.phantom.codes/v1
Key: your sk-...

// 12

Rate limits

// 13

FAQ