Phala Cloud Documentation — Private AI on TEE

Private AI Proxy verifies the Phala gateway on your machine and connects your agents to a local API. Agents receive a local token instead of your Phala credential. Tools can also connect to Phala directly.

Connect an Agent

Install Private AI Proxy and Sign In

Install the desktop app or CLI. Add a profile and choose Phala. Sign in with a Phala account or paste a key from Private AI → API Keys in the Phala dashboard.

Turn On Protected

The app verifies the gateway’s attestation and pins the channel before sending model traffic. Signed response receipts are audited after delivery.

Connect the Agent

On the Agents screen, connect the agent and restart it. The app writes its Local API into the agent’s provider settings; disconnecting restores the previous settings.

Agent	What Changes	Guide
Claude Code	Messages provider in `~/.claude/settings.json`	Claude Code
Codex	Custom Responses provider in `~/.codex/config.toml`	Codex CLI
DeepSeek Harness	`~/.dsh/cordis.patch.yml` and `.credentials.yaml`
OpenCode	Provider in `~/.config/opencode/opencode.json`	OpenCode
Pi	Provider in `~/.pi/agent/models.json`	Pi
Hermes	Provider in `~/.hermes/config.yaml`	Hermes
OpenClaw	Native-host provider; restart required	OpenClaw
Oh My Pi	Local token and models file; restart `omp`

From the CLI, sign in and start protection before connecting an agent:

pap profiles login work --provider phala
pap start --profile work
pap agents connect codex

See Private AI Proxy CLI for API-key profiles and other commands.

Other Tools

Tools with an OpenAI-compatible provider setting, such as Cline, can use the Proxy’s Local API while Protected is on:

Field	Value
Base URL	`http://127.0.0.1:4180/v1`
API Key	The client key from the app’s Overview screen
Model	`z-ai/glm-5.3`, or another chat model from the app’s verified catalog

See Other Tools for Continue, Zed, Copilot, Aider, Goose, and chat apps.

Without the Proxy

On servers, in CI, or in frameworks, point the client at Phala directly. Every model on this endpoint runs in a TEE, but a base-URL configuration does not verify attestation locally. Use the TypeScript ACI Verifier to enforce a verified channel in Node.js or Bun applications.

Client	Protocol	Base URL
OpenAI clients and agent frameworks	Chat Completions	`https://inference.phala.com/v1`
Codex CLI and other Responses clients	Responses	`https://inference.phala.com/v1`
Claude Code and Anthropic clients	Messages	`https://inference.phala.com` (the client appends `/v1/messages`)

For direct connections, create a key under Private AI → API Keys and export it in the client’s shell:

export PHALA_AI_API_KEY="YOUR_PHALA_API_KEY"
curl -s https://inference.phala.com/v1/models \
  | jq -r '.data[] | select((.supported_features // []) | index("tools")) | .id'

Use an exact chat-model ID such as z-ai/glm-5.3 from /v1/models. For SDKs, observability tools, and gateways, see Migration and Integrations.

​Connect an Agent

​Other Tools

​Without the Proxy

Connect an Agent

Other Tools

Without the Proxy