Mintplex-Labs/anything-llm · critical · Error

No Foundry Base Path was set.

Error message

No Foundry Base Path was set.

What it means

Thrown by the FoundryLLM constructor when process.env.FOUNDRY_BASE_PATH is unset. Foundry Local is Microsoft's on-device inference runtime that exposes an OpenAI-compatible HTTP endpoint; AnythingLLM needs that endpoint URL (later passed through parseFoundryBasePath) to construct its internal OpenAI client. Without a base path there is no server to route completions to, so the provider refuses to instantiate.

Source

Thrown at server/utils/AiProviders/foundry/index.js:30

const { OpenAI: OpenAIApi } = require("openai");
const ToolCallTextFilter = require("./toolCallFilter.js");

class FoundryLLM {
  /**
   * The largest context window we will select on the user's behalf.
   * Foundry runs on the user's own machine with no performance setting, so a
   * model advertising 128K would make an average laptop crawl. A user who
   * explicitly sets a larger limit still gets it, up to the model's real window.
   * @type {number}
   */
  static MAX_DEFAULT_CONTEXT_WINDOW = 16_000;

  /** @see FoundryLLM.cacheContextWindows */
  static modelContextWindows = {};

  constructor(embedder = null, modelPreference = null) {
    if (!process.env.FOUNDRY_BASE_PATH)
      throw new Error("No Foundry Base Path was set.");

    this.className = "FoundryLLM";
    this.model = modelPreference || process.env.FOUNDRY_MODEL_PREF;
    this.openai = new OpenAIApi({
      baseURL: parseFoundryBasePath(process.env.FOUNDRY_BASE_PATH),
      apiKey: null,
    });

    this.embedder = embedder ?? new NativeEmbedder();
    this.defaultTemp = 0.7;
    this.limits = null;
    FoundryLLM.cacheContextWindows(true);
    this.#log(`Loaded with model: ${this.model}`);
  }

  static #slog(text, ...args) {
    console.log(`\x1b[36m[FoundryLLM]\x1b[0m ${text}`, ...args);
  }

View on GitHub (pinned to 526360e320)

Solutions

1. Set FOUNDRY_BASE_PATH in .env (or the provider settings UI) to your Foundry Local OpenAI endpoint, e.g. http://localhost:5273/v1
2. Confirm Foundry Local is installed and running with `foundry --status` (or the tray app) before starting AnythingLLM
3. Restart the AnythingLLM server so the new environment variable is loaded into process.env
4. Verify the endpoint with `curl $FOUNDRY_BASE_PATH/models` to confirm it answers before configuring AnythingLLM

Example fix

// before (server/.env has no Foundry vars)
//   selecting Foundry Local in the UI throws on construction

// after
// server/.env
//   FOUNDRY_BASE_PATH=http://localhost:5273/v1
//   FOUNDRY_MODEL_PREF=Phi-3-mini-4k-instruct-cuda

// before constructing the provider
function canUseFoundry() {
  return Boolean(process.env.FOUNDRY_BASE_PATH);
}
if (!canUseFoundry()) {
  // surface a user-facing config error instead of letting the constructor throw
  throw new ConfigError('Set FOUNDRY_BASE_PATH to your Foundry Local endpoint.');
}

When it happens

Trigger: Constructing `new FoundryLLM(embedder, modelPreference)` at provider-selection time while FOUNDRY_BASE_PATH is absent from process.env. In AnythingLLM this fires the moment a user picks the Foundry Local LLM provider but never supplied the endpoint in the UI or .env.

Common situations: Foundry Local not yet installed or not running; the .env file has no FOUNDRY_BASE_PATH line; the user selected Foundry as the provider before finishing setup; a Docker/PM2 deployment that did not pass the env var through to the process.

Related errors

• KoboldCPP must have a valid base path to use for the api.
• Could not load ${this.model} into Foundry Local: ${error}
• No Gemini API key was set.
• GenericOpenAI must have a valid base path to use for the api
• No Gitee AI API key was set.

AI-assisted analysis of Mintplex-Labs/anything-llm@526360e320 (2026-08-13). Data as JSON: /api/errors/45e0054ed9fa3b0e. Report an issue: GitHub.