Skip to main content
Langdock is an enterprise AI workspace for chat, agents, and workflows. It supports OpenAI-compatible providers, so you can point it at Requesty and use any model in the Model Library from a single key, with Requesty analytics, approved-model lists, and fallback policies applied to every request. Setup has three parts: add a Requesty key to Langdock, add each model you want to expose as a custom model, and test it in chat.

​
1. Create a Requesty API key

  1. Open the API Keys page in the Requesty platform.
  2. Click Create Key, give it a name such as Langdock, and copy the key.
Optionally restrict which models this key can use with Approved Models or an Access List. Langdock only exposes the models you add manually in step 3, but limiting the key is good hygiene.

​
2. Add the key in Langdock

  1. In Langdock go to Settings > Models and scroll down to Keys.
  2. Click Add key.
Langdock Settings > Models with the Add key button highlighted
  1. Choose OpenAI Compatible as the provider type and fill in the form:
  1. Click Save. The key appears under Keys with the status Not used until a model references it.
Langdock OpenAI Compatible key form filled in with the Requesty base URL

​
3. Add models as custom models

Langdock does not fetch the model list from the provider. Every model you want to use across the organisation has to be added individually as a custom model.

​
3a. Open the custom model wizard

In Settings > Models, select the Custom tab and click Add custom model.
Custom tab in Langdock with the Add custom model button highlighted

​
3b. Set up the model manually

The wizard opens on Choose from configurations, which lists Langdock’s pre-configured providers. Ignore that list and click set up manually in the bottom right corner.
Langdock model wizard with the set up manually link highlighted

​
3c. Configure the model

Fill in the Configure step: Leave Visible to users, Available in agents, and Advanced capability enabled unless you want to restrict the model. Click Continue.
Langdock Configure step with model name and provider model name filled in
A bare name like deepseek-v4.1-flash is a Managed Policy: Requesty picks the best provider for that model on every request and handles fallbacks for you. Append @eu (e.g. deepseek-v4.1-flash@eu) to keep routing on EU-hosted endpoints. To pin a specific provider instead, use the provider/model-name ID from the Model Library, for example anthropic/claude-sonnet-4-6. Custom policies from your organisation work too with their policy/ prefix.

​
3d. Configure the deployment and test

On the Deployment step:
  1. Under API Key, select the Requesty key you created in step 2.
  2. Set Model Name / ID to the same Requesty model ID.
  3. Leave Tokens per minute limit empty unless you want Langdock to rate-limit this model.
  4. Click Test & continue. Langdock sends a test request through Requesty and moves to the Review step on success.
Langdock Deployment step with the Requesty key selected
Check the summary on the Review step and click Save model.
Langdock Review step showing the DeepSeek v4.1 Flash model routed through the Requesty key
Repeat this section for every model you want to offer. Each one reuses the same Requesty key, so you only manage a single credential.

​
4. Test in chat

Open a new chat in Langdock, pick the model from the model selector, and send a message. The response is served through Requesty and shows up in your Requesty logs and analytics.
Langdock chat using the DeepSeek v4.1 Flash model through Requesty

​
Troubleshooting

  • Test & continue fails: confirm the Model Name / ID matches an entry in the Model Library and that the key is allowed to use it under Approved Models.
  • 401 Unauthorized: the API key was pasted incorrectly or has been revoked. Edit the key under Keys and paste it again.
  • Model missing from the chat selector: make sure Visible to users is enabled on the model and that the deployment toggle is on.

​
Resources

Last modified on September 30, 2026