Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Go agentagenticaichatbotdsh-pluginembeddingsevaluationgenerative-aigolangknowledge-basellmmulti-tenantollamaopenaiquestion-answeringragrerankingsemantic-searchvector-searchwiki
import { get, post, put } from '@/utils/request'
export interface CatalogModel {
id: string
type?: string
match?: string
name?: string
context_window?: number
max_output_tokens?: number
reasoning?: boolean
deprecated?: boolean
[key: string]: unknown
}
export interface CatalogProvider {
id: string
name: string
api: string
model_types: string[]
settings?: Record<string, unknown>
models: CatalogModel[] | null
// Levels each chat model offers once thinking is on, resolved server-side.
model_thinking_levels?: Record<string, string[]>
vendor_thinking_levels?: string[]
}
export interface CatalogOverlay { providers: Record<string, Record<string, any>> }
export interface CatalogState {
version: number
baseline: string
overlay: CatalogOverlay
history: Array<{ version: number; overlay: CatalogOverlay; updated_by: string; updated_at: string }>
builtin: CatalogProvider[]
deployment: CatalogProvider[]
effective: CatalogProvider[]
deployment_error?: string
sync_error?: string
applied_version: number
}
// A check returns only the candidate; history and the lower layers are omitted.
export type CatalogPreview = Omit<CatalogState, 'history' | 'builtin' | 'deployment'>
export interface CatalogUpdate { version: number; baseline: string; overlay: CatalogOverlay }
const root = '/api/v1/system/admin/model-catalog'
export const getModelCatalog = () => get(root) as unknown as Promise<CatalogState>
export const previewModelCatalog = (request: CatalogUpdate) => post(`${root}/preview`, request) as unknown as Promise<CatalogPreview>
export const publishModelCatalog = (request: CatalogUpdate) => put(root, request) as unknown as Promise<CatalogState>