Phala Cloud Documentation — Confidential AI on TEE

Phala Confidential AI is OpenAI-compatible. Most integrations require changing only the base URL and API key:

https://inference.phala.com/v1

After migration, responses can be verified with x-receipt-id, Attestation Report, and Get Receipt.

Migrate from OpenAI

from openai import OpenAI

# Before
openai_client = OpenAI(api_key="sk-...")

# After
phala_client = OpenAI(
    api_key="YOUR_PHALA_API_KEY",
    base_url="https://inference.phala.com/v1",
)

Everything else stays familiar: chat completions, streaming, tool calling, vision, structured output, and embeddings use OpenAI-compatible request shapes.

Migrate from OpenRouter

from openai import OpenAI

client = OpenAI(
    base_url="https://inference.phala.com/v1",
    api_key="YOUR_PHALA_API_KEY",
)

Model ids use the provider/model format returned by /v1/models. Use is_tee: true to find models that can be served confidentially, then verify the actual response receipt.

OpenAI SDK

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_PHALA_API_KEY",
    base_url="https://inference.phala.com/v1",
)

response = client.chat.completions.create(
    model="z-ai/glm-5.2",
    messages=[{"role": "user", "content": "Explain TEE privacy."}],
)

print(response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.PHALA_AI_API_KEY,
  baseURL: "https://inference.phala.com/v1",
});

const response = await client.chat.completions.create({
  model: "z-ai/glm-5.2",
  messages: [{ role: "user", content: "Explain TEE privacy." }],
});

LangChain

from langchain_openai import ChatOpenAI

chat = ChatOpenAI(
    model="z-ai/glm-5.2",
    api_key="YOUR_PHALA_API_KEY",
    base_url="https://inference.phala.com/v1",
)

response = chat.invoke("Explain confidential inference in one paragraph.")
print(response.content)

Vercel AI SDK

import { generateText } from "ai";
import { createOpenAI } from "@ai-sdk/openai";

const phala = createOpenAI({
  apiKey: process.env.PHALA_AI_API_KEY,
  baseURL: "https://inference.phala.com/v1",
});

const { text } = await generateText({
  model: phala("z-ai/glm-5.2"),
  prompt: "Explain attestation.",
});

Langfuse

import os
from langfuse.openai import OpenAI

os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-..."
os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-..."
os.environ["LANGFUSE_HOST"] = "https://cloud.langfuse.com"

client = OpenAI(
    api_key="YOUR_PHALA_API_KEY",
    base_url="https://inference.phala.com/v1",
)

Observability tools may store prompts, responses, metadata, or traces. For sensitive data, configure redaction or self-hosting before sending production traffic.

For custom Node.js or Bun applications, use the TypeScript ACI Verifier to inject a pinned fetch and audit signed receipts in code.

Verified Coding Agents

Pi and OpenCode can use native Phala Cloud providers that verify the ACI workload and TLS channel before model traffic is sent, then verify each signed response receipt before the agent finishes the turn.

Pi

Install the provider through Pi’s package manager:

pi install npm:pi-provider-phala-cloud

Start Pi and use its native login and model picker:

/login phala
# choose Phala Cloud account or Phala Cloud API key
# wait for the footer to show aci-verified
/model
# search for phala/, select a model, and press Ctrl+S to save it as the default

Pi stores the credential, model catalog, and saved default through its native persistence. PHALA_AI_API_KEY is also supported for the current process.

OpenCode

Install the provider through OpenCode’s plugin command:

opencode plugin opencode-provider-phala-cloud --global

Restart OpenCode, then use its native provider and model pickers:

/connect
# select Phala Cloud and choose the account flow or API-key method
/models
# search for phala/ and select a model

OpenCode stores the plugin configuration and credential in its native stores. Do not add a separate provider.phala block. PHALA_AI_API_KEY is also supported for the current process. Both providers expose /phala-attestation, /phala-receipts, /phala-receipt, and /phala-session for inspecting verified evidence. These commands are optional: connection and response verification already fail closed automatically.

Changing only an OpenAI-compatible base URL does not inject ACI’s attested TLS transport. Base-URL clients can call the API and verify receipts separately; use the native Pi or OpenCode provider when verification must be enforced inside the agent.

Coding Assistants and Desktop Clients

For tools that support an OpenAI-compatible provider, use:

Setting	Value
Provider type	OpenAI compatible
Base URL	`https://inference.phala.com/v1`
API key	Your Phala API key
Model	Any id from `/v1/models`

This works for tools such as Continue, Cline-compatible assistants, Cherry Studio, and other clients that let you set an OpenAI-compatible base URL.

Migration Checklist

Create a Phala API key.
Change the base URL to https://inference.phala.com/v1.
List models with GET /v1/models.
Use is_tee: true for confidential-capable models.
Capture x-receipt-id in raw HTTP responses when you need verification.
Verify one representative response before production rollout.

Confidential AI

Migration and Integrations

Migrate from OpenAI

Migrate from OpenRouter

OpenAI SDK

LangChain

Vercel AI SDK

Langfuse

Verified Coding Agents

Pi

OpenCode

Coding Assistants and Desktop Clients

Migration Checklist

On-demand API

Error Handling

​Migrate from OpenAI

​Migrate from OpenRouter

​OpenAI SDK

​LangChain

​Vercel AI SDK

​Langfuse

​Verified Coding Agents

​Pi

​OpenCode

​Coding Assistants and Desktop Clients

​Migration Checklist

​Related

On-demand API

Error Handling

Migrate from OpenAI

Migrate from OpenRouter

OpenAI SDK

LangChain

Vercel AI SDK

Langfuse

Verified Coding Agents

Pi

OpenCode

Coding Assistants and Desktop Clients

Migration Checklist

Related