Skip to content

Admin Console

← User Guide · 简体中文

The admin console (admin-web) is for system administrators, deployed separately from the user web; sign in with an admin account. Sixteen pages across nine top-level menus: dashboard, users, model platforms, models, characters, preset characters, knowledge bases, MCP, workflow orchestration (components + flows), token monitoring, and system settings (storage / ASR / TTS / quota / rate limits).

Dashboard

Six stat cards: new users, new knowledge bases, new documents, characters, token consumption, image consumption — each showing the daily delta plus total / monthly consumption.

User Management

Search filters: name, status (pending activation / normal / disabled), created/updated time, admin flag.

Table: id, uuid, name, email, avatar, status, activation time, admin flag, etc.

New user form fields:

FieldDescription
Name (required), passwordCreate the account
Daily / monthly token quotaToken consumption caps
Daily / monthly request quotaLLM request caps
Daily / monthly image quotaImage generation caps

Row actions: Edit, Disable / Enable (disable freezes login). When activation emails go missing, set the user's status to normal to activate on behalf.

Model Platforms & Models

The admin's most-used feature — "empty model list" or "model unavailable" for users is usually solved here.

Platform Configuration

Table: name, title, base URL, API Key (configured / not), Secret Key, proxy flag, OpenAI-compatible flag.

New / edit fields:

FieldIn-app hint (translated)
Name / title (required), base URL, API key, secret keyPlatform basics; built-in options: Openai, Lingji, Qianfan, Ollama, DeepSeek, SiliconFlow
Enable proxyForm note: "Detailed proxy configuration lives in the project's configuration files"
OpenAI API compatibleForm note: "OpenAI-API-compatible platforms need no extra code adaptation" — chat-type models integrate with zero code

WARNING

Deleting a platform confirms "cannot be recovered and may affect model calls…"; built-in default platforms cannot be deleted.

Model Management

Toolbar: new, Refresh status (full health probe, "Health probe triggered").

Table: name, title, type, platform, enabled, health (🟢 normal / 🔴 abnormal / disabled), free flag, context window, max input/output tokens, etc.

Key form fields:

FieldDescription
Type (required)Text / image / embedding / rerank / TTS / ASR
Platform (required)Dropdown from platform configuration
Input types (required, multi)Text / image / audio / video
Output formats (required, multi)Text / JSON
Is reasoner / thinking closable / supports web searchDrives the user-side toggles
PropertiesJSON editor, e.g. TTS voices {"voices":[...]}, image sizes
Context window, max input/output tokensNumbers

Row actions: edit, enable/disable, set free / set paid, delete.

Integrating a brand-new platform (database config + optional code): Model Platform Integration Guide.

Characters & Presets

  • Characters: page banner reads "In this system a character is an AI persona"; browse all users' characters (title, persona, total tokens, context flag); edit/delete;
  • Preset characters: maintain the user-side preset library: title, type (11 categories), description, persona, knowledge base name (placeholder: "Leave empty to skip auto-creating a knowledge base" — when set, copying the preset auto-creates a same-named dedicated base).

Knowledge Base Management

Search: title, owner name, public flag, time.

Table: name, owner, document count, vector count, like count, public flag, plus the full configuration folded in a column: split overlap, split strategy, max tokens per segment, ingest model, max retrieval count, min retrieval score, LLM temperature, system prompt.

The edit dialog mirrors the user-side create form (custom-separator presets \n\n (paragraph), \n (newline); free input); edit only, no create.

MCP Management

Register / maintain MCP services. Edit form:

FieldIn-app hints (translated)
TransportSSE (tagged "deprecated" + alert: "SSE is deprecated by the new MCP spec; use Streamable HTTP for new services") / streamable-http / stdio
Install typedocker / local / remote / WebAssembly
SSE URL, SSE timeout (s, default 30) / STDIO command & argsShown per transport
ParametersColumns: parameter name ("must match the parameter name in the MCP Server"), parameter title ("display only, to convey meaning"), value, sensitive ("sensitive information is stored encrypted"); "+add parameter"
Parameter definitions"Definitions users fill with actual values later" — drives the user-side configure dialog
Website, description, enabled

Workflow Orchestration

  • Components: maintain canvas node components. Name hints: "1. The name is the component's unique identifier and cannot change after saving" / "2. Letters only (a-z, A-Z), no spaces"; display order: "components' order in the designer, ascending"; Start/End cannot be disabled or deleted;
  • Flows: all workflows' title, public flag, enabled flag, creator; edit / toggle / delete.

Token Monitoring

Read-only. Filters: username, platform, model name, source type, request time range.

Table: username, source type, platform, model, input tokens, output tokens, duration (ms), request time.

Source types: unknown, character chat, KB Q&A, KB indexing, workflow node, agent, long-term memory extraction, long-term memory analysis — useful to locate where quota goes (indexing and memory extraction consume tokens too).

System Settings

PageDescription
StorageShows the current location (local / Aliyun OSS). Local: "files stored on the server's disk"; OSS: fill in access key id / secret / bucket / endpoint, then save and enable
ASRPick an ASR model (platform follows), max recognition duration (s), max file size (KB)
TTSBanner (translated): "Synthesizer side — 1. Client: use the client's (e.g. browser) TTS, ignoring model/platform parameters, free. 2. Server: server-side synthesis via LLMs such as cosyvoice-v2, usually paid."; pick the side + TTS model
Quota configurationNote: "changes take effect within 10 minutes". Four dimensions: tokens (daily/monthly max), requests (daily/monthly count), images (daily/monthly count), chat (daily/monthly questions)
Rate limitsNote: "changes take effect within 10 minutes". Text requests and image generation each set "requests + time window (minutes)"

Previous: Prompt Store · Next: API Reference