Mintplex-Labs/anything-llm · warning · Error

OMLX:cacheContextWindows - ${res.statusText}

Error message

OMLX:cacheContextWindows - ${res.statusText}

What it means

Thrown inside the static cacheContextWindows fetch chain when the oMLX /v1/models endpoint returns a non-2xx HTTP status (res.statusText). It is immediately caught by the following .catch, which logs 'Error caching context windows' and swallows it — callers never observe this throw; it appears only in server logs. The effect is that modelContextWindows stays empty and promptWindowLimit falls back to defaults.

Source

Thrown at server/utils/AiProviders/omlx/index.js:90

    try {
      // Skip if we already have cached context windows and we're not forcing a refresh
      if (Object.keys(OMLXLLM.modelContextWindows).length > 0 && !force) return;

      const endpoint = new URL(
        parseOMLXBasePath(process.env.OMLX_LLM_BASE_PATH)
      );
      endpoint.pathname += "/models";
      await fetch(endpoint.toString(), {
        headers: {
          "Content-Type": "application/json",
          ...(process.env.OMLX_LLM_API_KEY
            ? { Authorization: `Bearer ${process.env.OMLX_LLM_API_KEY}` }
            : {}),
        },
      })
        .then((res) => {
          if (!res.ok)
            throw new Error(`OMLX:cacheContextWindows - ${res.statusText}`);
          return res.json();
        })
        .then(({ data: models }) => {
          models.forEach((model) => {
            // A model can omit max_model_len - cache the 16k fallback for it
            // so it is not later mistaken for a large-context model.
            if (!model?.max_model_len)
              return (OMLXLLM.modelContextWindows[model.id] = 16000);
            OMLXLLM.modelContextWindows[model.id] = Number(model.max_model_len);
          });
        })
        .catch((e) => {
          OMLXLLM.#slog(`Error caching context windows`, e);
          return;
        });

      OMLXLLM.#slog(`Context windows cached for all models!`);
    } catch (e) {

View on GitHub (pinned to 526360e320)

Solutions

1. curl the oMLX /v1/models endpoint with the same API key to see the real status.
2. Correct OMLX_LLM_BASE_PATH and OMLX_LLM_API_KEY in .env.
3. Restart the server so cacheContextWindows re-runs after the fix.
4. Optionally set OMLX_LLM_TOKEN_LIMIT manually so context sizing does not depend on the cache.

Example fix

// before
.then((res) => {
  if (!res.ok) throw new Error(`OMLX:cacheContextWindows - ${res.statusText}`);
  return res.json();
})

// after - include status code for actionable logs
.then((res) => {
  if (!res.ok) throw new Error(`OMLX:cacheContextWindows - HTTP ${res.status} ${res.statusText}`);
  return res.json();
})

const probeOmlxModels = async (basePath, apiKey) => {
  const res = await fetch(`${basePath.replace(/\/$/, '')}/v1/models`, {
    headers: apiKey ? { Authorization: `Bearer ${apiKey}` } : {},
  });
  if (!res.ok) throw new Error(`oMLX /models unhealthy: HTTP ${res.status}`);
  return res.json();
};
await probeOmlxModels(process.env.OMLX_LLM_BASE_PATH, process.env.OMLX_LLM_API_KEY);

When it happens

Trigger: oMLX /v1/models returns 401 (bad/missing OMLX_LLM_API_KEY), 404 (wrong base path / endpoint moved), 500, or a transient HTTP error during constructor-time cacheContextWindows(true).

Common situations: Wrong or expired OMLX_LLM_API_KEY; oMLX server partially up (inference works but /models errors); base path pointing at a non-oMLX server; network blip at startup; oMLX version with a changed /models schema.

Related errors

• Model capabilities for ${this.model} could not be retrieved
• ${e.message}
• No OMLX API Base Path was set.
• OMLX must have a valid model set.
• OpenRouter Failed to embed: ${error}

AI-assisted analysis of Mintplex-Labs/anything-llm@526360e320 (2026-08-13). Data as JSON: /api/errors/617e88e8d0d0fc7a. Report an issue: GitHub.