perf: LLM API Fails with "fetch failed" Error When Using Ollama

[akhileshthite]

When attempting to use local LLM services with Ollama, the browser throws a TypeError: fetch failed error.

This occurs because Node.js's built-in fetch (via undici) imposes default timeouts that are too restrictive for large language models, which can take significant time.

Error:

Error: Error invoking remote method 'lIm-chat': TypeError: fetch failed

Potential solution:

import { Agent } from 'undici';

const llmAgent = new Agent({
  headersTimeout: 0, // No timeout for headers
  bodyTimeout: 0     // No timeout for body
});

// Applied to all fetch requests
const response = await fetch(url, {
  method: 'POST',
  headers,
  body: JSON.stringify(data),
  dispatcher: llmAgent // Use custom agent
});

This allows large models to take unlimited time to respond without hitting artificial timeout limits, enabling reliable local LLM functionality with Ollama.

Hardware:

• M1 Mac 8 gb RAM

[akhileshthite]

Addressed here:
p2plabsxyz/peersky-browser#96

[RangerMauve]

A PR for this fix would be welcome. I'd suggest using the streaming API when possible in the meantime.