Model Configuration per Call Type
| Provider | Call Type | Model | Max Tokens | Temp | Budget £ | Cost in/out (USD/MTok) | Enabled | Notes | |
|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||
—
Anthropic Sonnet 4.6: $3/$15 · Haiku 4.5: $1/$5 · Perplexity Sonar tiers vary · Google = TBD
Changes take effect on the next V5 pipeline run. The n8n canvas reads this config via
GET /api/admin/config/llm-config at the top of the workflow, then injects
provider, model, max_tokens, and (for Perplexity) extra_params.search_domain_filter
into each provider node via expressions like
={{ $('Get LLM Config').item.json.config.credit_memo.model }}.
Perplexity rows enforce an FCA-defensible domain allowlist (research auditability). Edit it via the
Allowlist button — JSON array of bare domains, one per host (no https://, no paths).