Configure providers, models, and aliases

A provider is the connection from Orchestris to a model service. Its protocol is the request format that service accepts. Choose the service first, then confirm that its base URL and protocol match the provider's documentation.

Open Integrations → Providers at /providers. A server admin must select an organization before making changes. The Provider choice fills the usual base URL and streaming protocol; review both before saving.

1. 1

   Open the provider form

   At /providers, select Add provider.

2. 2

   Choose Provider

   Select one of the provider choices above. For OpenAI-compatible services, choose the protocol the service actually implements.

3. 3

   Name the connection

   Enter an organization-specific Name that distinguishes this credential and endpoint from other connections.

4. 4

   Confirm Base URL and Streaming protocol

   Keep the preset URL unless the provider requires a documented override. Enter the complete API base for OpenAI-compatible services.

5. 5

   Enter an API key

   Enter a credential intended for this deployment and organization. The Server requires a non-empty API Key before an active provider can be saved and tested. Never place credentials in the provider name or URL.

6. 6

   Save, then test

   Keep Active selected, add the provider, and use Test connection from its details. A connection test is necessary but does not replace model inference verification.

1. 1

   Open Add Models

   Open the provider and select its Models tab, then Add Models.

2. 2

   Try Discover catalog

   Select the models that should be enabled and choose Add models. Discovery can be unavailable or incomplete for some upstreams.

3. 3

   Use Manual entry when needed

   Enter the exact model name used by the provider service when discovery does not return it. Review the context length, capabilities, tokenizer, and pricing before treating automatically discovered values as confirmed.

4. 4

   Verify inference

   Use Verify for an enabled model. Confirm a successful response and review the reported token source and provider request details.

5. 5

   Review accounting fields

   If provider usage is absent, confirm the tokenizer assignment before relying on local estimates. Review pricing and capability flags before production use.

Open Integrations → Model Aliases at /model-aliases and select Add alias. Enter a lowercase Alias key, a user-facing Name, and one Base model. Clients can use the alias while an admin intentionally changes its backing model with Retarget.

1. 1

   Open Directory → Groups

   At /groups, create or select a group and open Model Access.

2. 2

   Grant a base model or alias

   Enable Has Access for the intended entry, set any Daily Token Quota and Daily Cost Quota ($), then apply the changes.

3. 3

   Add members

   Use the group's Members tab or the user's Groups tab at /users. No group is assigned automatically.

4. 4

   Confirm as the member

   Refresh the Chat model picker. It should contain only models and aliases granted through that member's groups.

5. 5

   Review use

   Use Reports → Usage at /usage after a real request to review token and cost attribution.

• Configure groups, access, and quotasUnderstand grants, quota layers, and API-key ownership.
• Manage users and invitationsAdd members to the groups that provide model access.
• Review usage and operational healthVerify provider, model, token, and cost reporting.

This advanced option is only for a reviewed provider that runs on a private network or the same machine as Orchestris. Provider URLs that resolve to loopback or private network addresses are blocked by default as a server-side request forgery (SSRF) defense.

An operator can enable private-address connections for the whole Server process. Managed-workspace customers must ask their operator to review this requirement.

ORCHESTRIS_SECURITY__ALLOW_PRIVATE_IPS=true