modelSpecs.list requires min 1 entry, preventing agents-only UX with addedEndpoints

[eric-chamberlain-tw]

What happened?

When using addedEndpoints: [agents] to show only the Agents endpoint in the UI, the Zod schema requires modelSpecs.list to have at least 1 entry (.min(1) in specsConfigSchema). This makes it impossible to configure an agents-only UX where no raw model specs are shown:

modelSpecs:
  enforce: true
  prioritize: true
  addedEndpoints:
    - agents
  list: []    # Fails validation: "Array must contain at least 1 element"

The addedEndpoints field was added in config v1.2.4 (PR #6650) specifically to keep endpoints like agents visible alongside model specs. However, the .min(1) constraint on list predates addedEndpoints and was never relaxed, so there's no valid config that shows only agents with no raw model specs.

Additionally, processModelSpecs in packages/data-schemas/src/app/specs.ts returns undefined when list is empty (line 39), which would discard addedEndpoints even if the schema check were bypassed.

Expected behavior

modelSpecs.list should accept an empty array (or be optional) when addedEndpoints is provided, allowing configurations like agents-only UX without needing a dummy spec entry.

Suggested fix

In packages/data-provider/src/models.ts:

// Before
list: z.array(tModelSpecSchema).min(1),

// After
list: z.array(tModelSpecSchema).default([]),

And update processModelSpecs in packages/data-schemas/src/app/specs.ts to not discard addedEndpoints when list is empty.

Version Information

Commit 9054ca9c1 (current main)

Steps to Reproduce

1. Configure librechat.yaml with modelSpecs.enforce: true, addedEndpoints: [agents], and list: []
2. Start LibreChat
3. Process exits with: Invalid custom config file... Array must contain at least 1 element

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

I think this is not intended, because it's meant to generally add endpoint options on top of model specs. Seems like this could be an independent feature, though!

[danny-avila]

You can restrict endpoints via .env but this does not lock down usage to just agents

https://www.librechat.ai/docs/configuration/dotenv#endpoints

so noting this as a feature request

[eric-chamberlain-tw]

We are using Agents as a hack, because we need to control access to models on a per user basis.

I have a PR, if you want it.