n8n-nodes-panda-free-llm

An n8n community node that sends a prompt to multiple free LLM providers with automatic failover. If the first provider is rate-limited, out of daily quota, busy, times out, or returns an empty answer, the node automatically tries the next one — and so on — until one succeeds.

Because every provider here exposes an OpenAI-compatible /chat/completions endpoint, the node uses a single request shape for all of them.

Supported free providers: Groq, Cerebras, Google Gemini (AI Studio), OpenRouter, Mistral.

How it works

You give the node a fixed System Prompt and a per-run User Prompt. It walks your provider list top-to-bottom and returns the first successful reply, plus an attempts log showing what happened with each provider.

output    → the model's text answer
provider  → which provider answered
model     → which model answered
usage     → token usage (if the provider returns it)
attempts  → per-provider log (great for debugging quotas)

Installation (self-hosted)

1. In n8n, go to Settings → Community Nodes → Install.
2. Enter the package name: n8n-nodes-panda-free-llm
3. Confirm, and the Panda Free LLM node appears in the nodes panel.

Unverified community nodes run on self-hosted n8n. They are not available on n8n Cloud unless verified by n8n.

Credentials

Create one Panda Free LLM API credential and paste in only the keys you have (blank providers are skipped automatically):

Usage

• System Prompt — your fixed instructions (set once).
• User Prompt — defaults to ={{ $json.text }}. Map it to whatever field carries your input, e.g. ={{ $json.message }}.
• Response Format — Text (plain text) or JSON. In JSON mode the node sends response_format: { type: "json_object" }, nudges the prompt to return JSON, and also returns the parsed object in outputParsed (handy for feeding straight into a database).
• Provider — the provider to try first.
• Model — a live dropdown of models for the selected provider (reopen it after changing the provider to refresh). OpenRouter is filtered to :free models. Leave blank to use the provider's default.
• Enable Failover + Fallback Models (in order) — if the primary fails (rate limit, quota, busy, timeout, empty), the node tries each fallback in order. Each fallback row is a single Provider — Model dropdown from the live list, so you can pick a specific model per fallback (including the same provider with a different model). Exact duplicates of the primary are skipped automatically.
• Options — Max Tokens, Temperature, Timeout (ms), and "Throw If All Providers Fail" (turn off to output success:false instead of erroring).

Output fields: output (text), provider, model, usage, attempts, and — in JSON mode — outputParsed.

JSON mode relies on each provider's OpenAI-compatible JSON support, which varies by model. If a model doesn't honor it, the prompt instruction still pushes it toward JSON, and a provider that rejects response_format simply fails over to the next.

License

MIT