Agents

Claude Desktop

Native Claude Desktop on 3P; not claude.ai, MCP or Claude Code.

Setting up with an AI assistant? Copy the setup prompt or open this guide as Markdown.

Setup prompt

Read https://zrelay.net/docs/ai.md and https://zrelay.net/docs/md/agents/coding-agents/claude-desktop.md first.

Configure Claude Desktop on 3P for Zrelay. Detect my OS, installed app version and whether configuration is managed. Read https://claude.com/docs/third-party/claude-desktop/in-app-configuration, https://claude.com/docs/third-party/claude-desktop/gateway and https://keys.zrelay.net/api/catalog. Preserve existing settings and never bypass managed policies. Use the documented in-app local configuration: Gateway, Static API key, base URL https://api.zrelay.net (without /v1), Bearer authentication. Let me enter my Zrelay key privately in the app; never request, reveal, read back, log or embed it in a prompt or command. Select an enabled Claude model for my key's pool and /v1/messages; claude-sonnet-5 in flex is an example, subject to the current catalog. Check one-hour caching and token limits before claiming compatibility. Add the explicit model in Models. Apply Changes, Save & Restart, then verify the selected gateway and model without inference. Do not edit Claude Code settings or MCP configuration for this purpose. If GUI automation is unavailable, give exact manual steps. Ask for consent before installation, Test connection, or any paid inference. Report configured versus untested capabilities; never claim streaming/tool loops were verified without testing them.

Manual setup

1. In installed Claude Desktop open Help > Troubleshooting > Enable Developer Mode, then Developer > Configure Third-Party Inference. Windows has the menu on the sign-in screen; macOS uses its system menu. The documented local route starts at sign-in without creating an Anthropic account. Stop if managed policy makes settings read-only.

2. Connection: Inference provider = Gateway; Credential kind = Static API key; Gateway base URL = https://api.zrelay.net; Gateway auth scheme = Bearer. Enter the key yourself in Gateway API key. Preserve existing settings.

3. In Models add claude-sonnet-5 explicitly for a matching Claude/Flex key and /v1/messages. Use Apply Changes > Save & Restart, then the available 3P/gateway entry. Inspect provider, URL and model without pressing Test connection or starting a conversation.

4. If diagnostics are needed, inspect only provider-status.txt and deployment-mode.txt locally, without publishing the complete report. Do not edit claude_desktop_config.json or ~/.claude/settings.json as a substitute for inference configuration.

Configuration and commands

Use the exact fields in the manual above in the client’s supported settings UI. This client has no file-based recipe here; preserve existing connections and enter the key locally.

Alternative: ask an AI assistant

Copy this complete prompt into your coding assistant. It describes configuration work; installation and paid tests need your authorization.

Setup prompt

Configure Claude Desktop on 3P for Zrelay. Detect my OS, installed app version and whether configuration is managed. Read https://claude.com/docs/third-party/claude-desktop/in-app-configuration, https://claude.com/docs/third-party/claude-desktop/gateway and https://keys.zrelay.net/api/catalog. Preserve existing settings and never bypass managed policies. Use the documented in-app local configuration: Gateway, Static API key, base URL https://api.zrelay.net (without /v1), Bearer authentication. Let me enter my Zrelay key privately in the app; never request, reveal, read back, log or embed it in a prompt or command. Select an enabled Claude model for my key's pool and /v1/messages; claude-sonnet-5 in flex is an example, subject to the current catalog. Check one-hour caching and token limits before claiming compatibility. Add the explicit model in Models. Apply Changes, Save & Restart, then verify the selected gateway and model without inference. Do not edit Claude Code settings or MCP configuration for this purpose. If GUI automation is unavailable, give exact manual steps. Ask for consent before installation, Test connection, or any paid inference. Report configured versus untested capabilities; never claim streaming/tool loops were verified without testing them.

Limits and verification

Chat/Cowork expects one-hour caching: the reviewed claude-sonnet-5 Flex route has it; Prime does not. Limits are 128000/4096. A Desktop request for 16384 output will fail, and no official budget control was established. This documents 3P configuration without promising end-to-end operation, 1M context, Enterprise-plan requirements or full tool parity.

These recipes were reviewed against documentation and public catalog data. No client installation, authenticated request, paid inference, streaming or tool loop was executed. Validate configuration first; a later short text test does not establish complete agent compatibility.

Recheck the live catalog and your key’s provider/pool. The reviewed examples have 128000 input and 4096 output maxima unless a client limitation above prevents controlling the request. Never print, log or commit credentials.