Claude Cowork

Route Claude Desktop (Cowork) through Requesty to access 300+ models with a single gateway

Claude Cowork is Anthropic’s desktop app for working with Claude. It now ships with a built-in Third-Party Inference mode. Once enabled, Claude Cowork sends every request to the Anthropic-compatible gateway of your choice instead of Anthropic’s cloud.

Point it at Requesty and you get:

• 300+ models available directly from the Claude Cowork model picker: Sonnet, Opus, Haiku, GPT, Gemini, Mistral, Llama, and more.
• Policies for Fallback, Load Balancing, and Latency Routing without changing a thing on the client.
• Spend tracking: every Cowork session shows up in usage analytics and cost tracking, tagged as Claude Cowork traffic via the X-Title header.
• No Anthropic account required. “Continue with Gateway” lets teams use Claude Cowork with just a Requesty key.
• Enterprise deployment: export a .mobileconfig (macOS) or .reg (Windows) file from the Cowork setup UI to roll the Requesty configuration out via MDM / Group Policy.

How it works

Prerequisites

• Claude Cowork installed. Download from Anthropic.
• A Requesty API key from the API Keys page.

Configuration

1. Enable Developer Mode

   Open Claude Cowork (no sign-in required) and go to Help > Troubleshooting > Enable Developer Mode.

   

   A new Developer menu appears in the menu bar.

2. Open the third-party inference panel

   From the new menu bar item, choose Developer > Configure Third-Party Inference.

   
3. Select Gateway and fill in the Requesty credentials

   In the Configure third-party inference panel, pick Gateway (Anthropic-compatible) as the inference backend and enter:

   Field	Value
   Gateway base URL	https://router.requesty.ai
   Gateway API key	Your Requesty API key
   Gateway auth scheme	bearer
   Gateway extra headers	X-Title: Claude-Cowork

   

   Why X-Title: Claude-Cowork? Requesty reads the X-Title header to tag requests in analytics. Setting it to Claude-Cowork lets you filter Cowork traffic separately from your API, Claude Code, or Cline usage in Usage Analytics and Cost Tracking. You can use any value you like, for example Claude-Cowork-Design-Team, to segment further.

   Click Apply locally to save. Use Export if you want a .mobileconfig for macOS MDM or a .reg file for Windows Group Policy / Intune (see MDM deployment below).

   The other three backends (Bedrock, Vertex, Foundry) talk directly to AWS / GCP / Azure and bypass Requesty. Use Gateway if you want routing, fallbacks, spend tracking, and access to non-Anthropic models.

4. Restart Claude Cowork and continue with Gateway

   Quit and reopen Claude Cowork. On launch you’ll see a chooser. Pick Continue with Gateway (labeled Local configuration). No Anthropic account is needed.

   
5. Pick a model or policy

   The model picker in the top-right now lists every model and policy your Requesty org has access to. Select one and start working.

   

Model selection

Any entry from the Model Library works. Model IDs follow the familiar provider/model-name format:

Provider	Example
Anthropic	anthropic/claude-sonnet-4-5
Anthropic (1M context)	anthropic/claude-opus-4-6[1m]
OpenAI	openai/gpt-4o
Google	google/gemini-2.5-pro
AWS Bedrock	bedrock/claude-opus-4-6
Mistral	mistral/mistral-large-latest

Policies follow policy/<policy-name>:

• policy/reliable-sonnet-4-5
• policy/opus-4-6
• policy/gemini-3-pro

Policies are the recommended default for Cowork. A single policy gives you automatic fallback, load balancing, and latency-aware routing without the user ever needing to pick a specific model.

To use the 1M token context window for Opus 4.6 or Sonnet 4.5, append [1m] to the model name, for example anthropic/claude-opus-4-6[1m].

EU data residency

For EU-only routing, change the gateway base URL to the EU router:

https://router.eu.requesty.ai

Everything else (auth scheme, headers, model IDs) stays the same. See EU Routing for the full list of EU-approved models.

MDM deployment

The Export button in the Configure third-party inference panel produces a ready-to-deploy configuration file so you can roll Requesty out to an entire team:

• macOS: .mobileconfig payload you can push with Jamf, Intune, Kandji, or any Apple MDM.
• Windows: .reg file suitable for Group Policy Preferences or Intune custom profiles.

Recommended payload for a fleet rollout:

Gateway base URL        https://router.requesty.ai
Gateway auth scheme     bearer
Gateway extra headers   X-Title: Claude-Cowork
Gateway API key         <per-user service-account key>

Issue one Requesty key per user (or per team) from the API Keys page or the Key Management API so you can attribute spend, set per-user limits, and revoke individual keys without disturbing the fleet.

Benefits of Requesty + Claude Cowork

300+ Models, One App

Sonnet, Opus, GPT, Gemini, Mistral, Llama, all selectable from the Cowork picker.

Policy-Driven Routing

Point Cowork at a policy and get fallback + load balancing without touching the client.

Spend Visibility

Filter Cowork traffic in analytics via the X-Title: Claude-Cowork tag.

MDM-Ready

Export .mobileconfig / .reg once and push the setup to your whole org.

Troubleshooting

Still seeing “Sign in to Anthropic” on launch

Claude Cowork only shows Continue with Gateway after a valid configuration has been saved. Re-open Developer > Configure Third-Party Inference, verify the four fields, and click Apply locally again. A full app restart is required.

401 Unauthorized from the gateway

• Confirm Gateway auth scheme is set to bearer (not x-api-key, which is reserved for URLs ending in *.anthropic.com).
• Re-paste your Requesty API key. Trailing whitespace from the clipboard is a common culprit.
• Verify the key is active at app.requesty.ai/api-keys.

Model isn’t in the picker

Cowork pulls the model list through the gateway, so the model must be approved for your Requesty org. Check:

• The model exists in the Model Library.
• It’s in your org’s Approved Models list (if your org restricts models).
• For policies, the policy name must match policy/<name> exactly.

Cowork traffic isn’t tagged in analytics

Open Developer > Configure Third-Party Inference and confirm the Gateway extra headers field contains X-Title: Claude-Cowork (or your custom value). The header is sent on every request and surfaces in Usage Analytics as a filterable field.

Developer menu doesn’t appear

Help > Troubleshooting > Enable Developer Mode is a toggle. If the menu isn’t showing, it may have been disabled by an MDM profile. Ask your admin, or deploy the exported .mobileconfig directly so end users don’t need Developer Mode at all.

Reset the configuration

Help > Troubleshooting > Reset App Data wipes the Cowork configuration. Useful when switching between gateway and Anthropic cloud mode during testing.

Resources

• Claude Cowork product page
• Anthropic’s official Cowork third-party setup guide
• Requesty API Keys
• Requesty Model Library
• Fallback Policies, Load Balancing, Latency Routing
• EU Routing