mcp-model-relay

// src/config.ts
import { createConfigSchematics } from "@lmstudio/sdk";

export const globalConfigSchematics = createConfigSchematics()
  /* Main remote provider */
  .field(
    "mainApiKey",
    "string",
    {
      displayName: "Main Provider API Key",
      isProtected: true,
      placeholder: "sk-...",
    },
    "",
  )
  .field(
    "mainBaseUrl",
    "string",
    {
      displayName: "Main Provider Base URL",
      subtitle: "OpenAI-compatible endpoint for the final reasoning model.",
      placeholder: "https://api.deepseek.com/v1",
    },
    "https://api.deepseek.com/v1",
  )
  .field(
    "mainModel",
    "string",
    {
      displayName: "Main Model",
      subtitle: "Remote/main model used for the final analysis.",
      placeholder: "deepseek-v4-flash",
    },
    "deepseek-v4-flash",
  )

  /* Main model generation */
  .field(
    "mainTemperature",
    "numeric",
    {
      displayName: "Main Temperature",
      subtitle: "Use -1 to omit/provider default.",
    },
    0.1,
  )
  .field(
    "mainTopP",
    "numeric",
    {
      displayName: "Main Top P",
      subtitle: "Use -1 to omit/provider default.",
    },
    -1,
  )
  .field(
    "mainMaxTokens",
    "numeric",
    {
      displayName: "Main Max Tokens",
      subtitle: "Use 0 to omit/provider default.",
    },
    4000,
  )
  .field(
    "mainReasoningEffort",
    "string",
    {
      displayName: "Main Reasoning Effort",
      subtitle: "Provider-specific. Use provider_default to omit.",
      placeholder: "provider_default",
    },
    "provider_default",
  )
  .field(
    "mainReasoningDisplay",
    "string",
    {
      displayName: "Main Reasoning Display",
      subtitle: "How to display reasoning_content from the main model: hidden, status, or inline.",
      placeholder: "status",
    },
    "status",
  )
  .field(
    "deepSeekThinking",
    "string",
    {
      displayName: "DeepSeek Thinking",
      subtitle: "Use disabled for MCP/tool workflows unless reasoning_content is round-tripped.",
      placeholder: "disabled",
    },
    "disabled",
  )

  /* Local LM Studio orchestrator */
  .field(
    "localBaseUrl",
    "string",
    {
      displayName: "Local LM Studio Base URL",
      subtitle: "Native LM Studio REST API base URL. Do not include /api/v1/chat.",
      placeholder: "http://localhost:1234",
    },
    "http://localhost:1234",
  )
  .field(
    "localApiToken",
    "string",
    {
      displayName: "Local LM Studio API Token",
      subtitle: "Required only if LM Studio authentication is enabled.",
      isProtected: true,
      placeholder: "lm-studio",
    },
    "",
  )
  .field(
    "localModel",
    "string",
    {
      displayName: "Local Orchestrator Model",
      subtitle: "Local model used only for chat cleanup, intent extraction, and task preparation.",
      placeholder: "mistral-nemo-instruct-2407",
    },
    "",
  )
  .field(
    "localContextLength",
    "numeric",
    {
      displayName: "Local Context Length",
      subtitle: "context_length passed to LM Studio /api/v1/chat.",
    },
    8000,
  )
  .field(
    "localMaxOutputTokens",
    "numeric",
    {
      displayName: "Local Max Output Tokens",
      subtitle: "Maximum output for the local chat summary. Tool outputs are preserved separately by the relay.",
    },
    1000,
  )
  .field(
    "localTemperature",
    "numeric",
    {
      displayName: "Local Temperature",
      subtitle: "Use 0 for deterministic compression/planning.",
    },
    0,
  )
  .field(
    "localIntegrations",
    "string",
    {
      displayName: "Local MCP Integrations",
      subtitle: "Optional. Usually leave empty. The local model should not process raw tool output.",
      placeholder: "mcp/re-memory",
    },
    "",
  )
  .field(
    "localAllowedTools",
    "string",
    {
      displayName: "Local Allowed Tools",
      subtitle: "Comma-separated tool names for local integrations. Leave empty to expose all tools from listed integrations.",
      placeholder: "re_memory_search,re_snapshot_load_latest",
    },
    "",
  )
  .field(
    "localReasoning",
    "string",
    {
      displayName: "Local Reasoning",
      subtitle: "LM Studio /api/v1/chat reasoning setting. Use provider_default to omit.",
      placeholder: "provider_default",
    },
    "provider_default",
  )
  .field(
    "localOrchestratorTimeoutMs",
    "numeric",
    {
      displayName: "Local Orchestrator Timeout (ms)",
      subtitle: "Abort the local LM Studio orchestrator call after this time. Use 0 to disable.",
    },
    60000,
  )

  /* Prompt / relay limits */
  .field(
    "maxMessages",
    "numeric",
    {
      displayName: "Max Messages",
      subtitle: "Maximum recent chat messages included in the relay window. 0 = all.",
    },
    30,
  )
  .field(
    "maxToolResultChars",
    "numeric",
    {
      displayName: "Max Tool Output Chars",
      subtitle: "Maximum chars kept per raw tool output before deterministic relay truncation.",
    },
    30000,
  )
  .field(
    "maxPromptChars",
    "numeric",
    {
      displayName: "Max Prompt Chars",
      subtitle: "Abort if the direct prompt becomes too large before orchestration.",
    },
    200000,
  )
  .field(
    "failOpen",
    "boolean",
    {
      displayName: "Fail Open",
      subtitle: "If local orchestration fails, call the main provider directly.",
    },
    true,
  )
  .field(
    "showRelayStats",
    "boolean",
    {
      displayName: "Show Relay Stats",
      subtitle: "Append local/main model size stats after each response.",
    },
    true,
  )
  .field(
    "alwaysAvailableToolPrefixes",
    "string",
    {
      displayName: "Always Available Tool Prefixes",
      subtitle: "Comma-separated tool name prefixes that are never removed by completed-action filtering.",
      placeholder: "re_memory_,re_snapshot_",
    },
    "re_memory_,re_snapshot_",
  )
  .build();

export const configSchematics = createConfigSchematics()
  .field(
    "mainModelOverride",
    "string",
    {
      displayName: "Main Model Override",
      subtitle: "Optional. Leave empty to use global main model.",
      placeholder: "deepseek-v4-pro",
    },
    "",
  )
  .field(
    "localModelOverride",
    "string",
    {
      displayName: "Local Model Override",
      subtitle: "Optional. Leave empty to use global local orchestrator model.",
      placeholder: "mistral-nemo-instruct-2407",
    },
    "",
  )
  .field(
    "bypassLocalOrchestrator",
    "boolean",
    {
      displayName: "Bypass Local Orchestrator",
      subtitle: "Call the main provider directly for this chat.",
    },
    false,
  )
  .field(
    "debug",
    "boolean",
    {
      displayName: "Debug Logging",
      subtitle: "Log local and main request metadata.",
    },
    false,
  )
  .build();