LLM API settings

Connect a tenant-approved model provider for import mapping, validation explanations, and future AI workflows.

API keys are not stored in this demo. Production must encrypt keys, restrict tenant access, audit usage, and require approval before sending import samples to a model.

Gate 6 integration readiness

Only redacted headers and a small sample may be sent after explicit tenant approval.

Local assistant only / live model calls disabled

Encrypted key storage, tenant approval, prompt retention policy, and provider API clients are not configured.

Provider registry exists: complete Custom endpoints must use HTTPS: complete Secrets are masked in UI/test payloads: complete Raw import files blocked from model prompts: complete Live API clients and encrypted key storage: blocked_until_credentials

Add LLM API Provider

Provider

Connection name

Model or deployment

Endpoint URL

API key

Provider Options

OpenAI

ready_for_api_key

Use for header mapping suggestions, validation explanations, and import summaries after user approval.

Models: gpt-4.1-mini, gpt-4.1

Azure OpenAI

ready_for_endpoint_and_key

Requires tenant-specific Azure endpoint, deployment name, and API key.

Models: deployment_name

Anthropic

ready_for_api_key

Use only after import data sharing and retention rules are confirmed.

Models: claude-sonnet-4-5

Google Gemini

ready_for_api_key

Useful for mapping assistance with explicit tenant approval.

Models: gemini-2.5-flash

Custom compatible endpoint

requires_https_endpoint

Endpoint must be HTTPS and controlled by the tenant or trusted provider.

Models: custom-model

Provider contract coverage

Provider	Contract status	Data policy	Raw import allowed	Live enabled
OpenAI	guarded_contract_ready	redacted_headers_and_sample_rows_only_after_approval	No	No
Azure OpenAI	guarded_contract_ready	redacted_headers_and_sample_rows_only_after_approval	No	No
Anthropic	guarded_contract_ready	redacted_headers_and_sample_rows_only_after_approval	No	No
Google Gemini	guarded_contract_ready	redacted_headers_and_sample_rows_only_after_approval	No	No
Custom compatible endpoint	guarded_contract_ready	redacted_headers_and_sample_rows_only_after_approval	No	No

Connection Slots

Import mapping assistant

OpenAI · gpt-4.1-mini

sandbox_ready

Enterprise import reviewer

Azure OpenAI · tenant-deployment-name

endpoint_required

Quote wording reviewer

Anthropic · claude-sonnet-4-5

owner_approval_required

Private model endpoint

Custom compatible endpoint · custom-model

https_endpoint_required

Import AI Guardrails

Do not send full imported files to an LLM by default.

Use headers and a small redacted sample for mapping suggestions only after user approval.

Never include API keys, passwords, payment card data, or private notes in prompts.

Store prompt and response hashes for audit instead of raw customer data.

Require human review before applying model-suggested mappings or row fixes.