Skip to main content
OpenClaw is an open source personal AI assistant that runs on your own devices and connects to the messaging channels you already use, such as WhatsApp, Telegram, Slack, Discord, Signal, and iMessage. Requesty is an official provider in OpenClaw. The Requesty plugin ships with OpenClaw and updates with it, so there is nothing extra to install. Pick Requesty in the onboarding wizard, paste your API key, and OpenClaw lists the models your key can use. Using the Requesty integration, you can:
  • Access 600+ models from Anthropic, OpenAI, Google, Mistral, and many other providers through one API key.
  • Pick any managed model from the OpenClaw model picker, with live pricing and context windows.
  • Track and manage your spend in one place.
  • Apply fallback policies and load balancing so your assistant keeps answering when a provider is down.
  • Keep all processing and inference in the EU with the EU endpoint.
OpenClaw dashboard chat with claude-haiku-5-5 selected, answering a request to write a haiku about OpenClaw and Requesty

Prerequisites

  • A Requesty API key from the API Keys page.
  • macOS, Linux, or Windows. The OpenClaw installer sets up Node.js for you if needed.

Setup

1

Install OpenClaw

On Windows, run iwr -useb https://openclaw.ai/install.ps1 | iex in PowerShell instead.The installer installs the openclaw CLI and starts the onboarding wizard. If you already have OpenClaw, run openclaw onboard to start the wizard.
2

Choose Requesty as your provider

Choose Quick start. At Model/auth provider, select More…, type requesty, and press Enter on Requesty.
OpenClaw onboarding provider search with requesty typed and Requesty, Unified LLM Gateway, as the only match
To skip the provider list, start the wizard with Requesty already selected:
3

Paste your API key

Paste your Requesty API key at Enter Requesty API key. OpenClaw sends a test request and shows Inference verified once the key works. The default model is claude-sonnet-5-5.
OpenClaw onboarding with a masked Requesty API key and an Inference ready box reading Inference verified: requesty/claude-sonnet-5-5
Finish the remaining wizard steps (channels, skills, and memory) as you like. They do not affect Requesty.
4

Pick a model in the dashboard

Open the dashboard:
Click the model name in the chat composer and search for a model. The list contains every Requesty managed model. Select one, for example claude-haiku-5-5.
OpenClaw dashboard model picker filtered by haiku, listing claude-haiku-5-5 and claude-haiku-4-5 from Requesty
5

Chat

Send a message. It goes through Requesty and shows up in your analytics dashboard. OpenClaw identifies itself with the X-Title: OpenClaw header, so you can tell its traffic apart from your other apps.

Non-interactive setup

For servers and scripts, pass the key on the command line:
OpenClaw also picks up REQUESTY_API_KEY from the environment. If the gateway runs as a background service, put the key in ~/.openclaw/.env so the service can read it.

Choosing models

Once a key is configured, OpenClaw fetches GET /v1/models/managed and lists every managed model, with Requesty’s live prices, context window, and vision and reasoning flags. Models are addressed by their bare managed name, such as claude-sonnet-5-5, gpt-6-luna, or gemini-3.8-flash, and Requesty picks the best hosting provider for each request. The list is the full public catalog and is not filtered by your organization’s approved models. If you pick a model your organization has not approved, the request fails. List them from the terminal, or with /models requesty in a chat:
Change the default model:
Switch for a single chat with /model requesty/claude-haiku-5-5.

Routing policies and pinned models

The model picker only lists managed models. To use one of your own routing policies, declare it under models.providers.requesty.models in ~/.openclaw/openclaw.json, then reference it as requesty/policy/<your-policy-name>:
The policy then appears in the model picker next to the managed models. Because the policy resolves on Requesty’s side, you can change its models in the Requesty console without touching OpenClaw.
OpenClaw reads everything after an @ in a model reference as the name of an auth profile. Model IDs that contain @, such as the managed claude-haiku-5-5@eu or the pinned bedrock/claude-haiku-5-5@eu-central-1, reach Requesty without the suffix and run on the default route. Put those models in a routing policy and reference the policy instead.

Full EU inference

To keep your OpenClaw traffic fully in the EU, both Requesty’s processing and the model inference must stay in the EU. See EU Routing for how the two layers work.
1

Point OpenClaw at the EU endpoint

The change applies without restarting the gateway. Requesty now processes and stores every OpenClaw request in Frankfurt, and the model list is fetched from the EU endpoint too.
2

Create an EU-only routing policy

The EU endpoint alone does not keep inference in the EU: a managed model like claude-haiku-5-5 can still run on a provider outside the EU. Because OpenClaw drops @eu suffixes (see the warning above), use a routing policy that contains only EU-hosted models.Create a Fallback Policy in the Requesty console, for example eu-haiku:Find more EU-hosted models by filtering the Model Library by EU regions.
3

Use the policy in OpenClaw

Declare the policy and make it the default model:
Your ~/.openclaw/openclaw.json now contains:
Haiku 5.5 EU now appears in the dashboard model picker, and requests that use it run on EU-hosted models.The picker still lists the global managed models, and those can run outside the EU. To make sure no one in your organization can pick one, also restrict the model hosting location in the next step.
4

Enforce EU for the whole organization

To make sure no OpenClaw install or other client can send traffic outside the EU, go to Compliance as an organization admin:
  • Restrict Serving region to EU, so Requesty rejects requests that do not use router.eu.requesty.ai.
  • Restrict Model hosting location to EU, so Requesty rejects requests for models hosted outside the EU.
See the Compliance Report for details.

Troubleshooting

Authentication failed during onboarding

Check that you pasted the full key from the API Keys page and that it has not expired. Test it directly:
If this works but OpenClaw does not, run openclaw onboard --auth-choice requesty-api-key again to replace the stored key.

A model is missing from the picker

The list shows Requesty’s managed models only. Routing policies and @ model IDs are never listed automatically, see Routing policies and pinned models.

The gateway service cannot find the key

When the gateway runs as a background service, it does not see variables exported in your shell. Put REQUESTY_API_KEY in ~/.openclaw/.env, or run onboarding so OpenClaw stores the key itself.

Resources

Last modified on November 19, 2026