- Generations: single-prompt text completion against a named model (Salesforce-hosted or your BYO LLM configuration)
- Chat generations: multi-turn conversation with system/user/assistant messages, same model selection
- Embeddings: vector embeddings for retrieval and similarity on your own data
- Feedback: rate a generation so the Trust Layer audit trail and model tuning can use it
- 4 routes (all write) on api.salesforce.com, Trust Layer masking, toxicity scoring and audit applied by Salesforce, documented in the AI integration guide (llms.txt)
Salesforce Models API (Einstein) Connector
Salesforce Models API — Einstein's REST access to the large language models configured in your org through the Einstein Trust Layer: text generations, chat generations and embeddings by model name, plus a feedback endpoint — registered in Apivom Atlas as 4 routes on api.salesforce.com/einstein/platform/v1.
What you can do
Use cases
- Governed generation inside Salesforce — Send prompts through the org's Trust Layer so masking, audit and zero-retention terms apply even when the caller is an external system.
- Embeddings for CRM retrieval — Embed case notes, knowledge articles or product descriptions with an org-approved model and search them from your own application.
- Feedback-driven tuning — Return end-user ratings on each generation so prompt templates and model choices in Einstein Studio can be adjusted with real signal.
The Models API lets code outside Salesforce use the same models the Einstein 1 platform uses inside it — the Salesforce-hosted models and any provider you have configured under Einstein Studio — with the Einstein Trust Layer in the path: data masking, toxicity detection, zero-retention agreements with the providers and an audit trail in your org. The four routes are small and stable: generations, chat generations and embeddings, each addressed by model name, and feedback. Apivom Atlas registers them on api.salesforce.com with a Connected App that carries the Einstein API scope. No Apivom product calls the Models API: Apivom's AI products run on Apivom's own model stack, and no workflow routes prompts through Salesforce. The connector is catalogued for customers whose AI governance requires that generation happens inside their Salesforce Trust Layer, so an Apivom workflow could be scoped to call it, per customer.
A Connected App on your Salesforce org (consumer key + secret) and your My Domain instance URL. The caller runs the OAuth 2.0 flow and sends the access token as Bearer; the instance URL becomes the per-request base URL. Apivom Atlas forwards, logs and stores nothing.
Frequently asked questions
Can I switch this connector on today?
The connector is registered in Apivom Atlas — every route is catalogued, the credential contract is published and the AI integration guide (llms.txt) is live — and your credential can be stored in Apivom Iris. Wiring it into a product workflow (sync, automation, notifications) is done per customer: tell us the use case and we scope it with you.
Do Apivom's AI products use the Salesforce Models API?
No. Apivom's AI products use Apivom's own model stack. This connector is registered for customers who need generation to stay inside their Salesforce Trust Layer; wiring an Apivom workflow to it is scoped per customer.
Which models can I call and who pays?
The models enabled in your org's Einstein Studio — Salesforce-hosted ones and any bring-your-own provider you configured. Usage is metered by Salesforce as Einstein Requests on your org; Apivom adds nothing.
Does Apivom charge per API call for this connector?
No. Connectors are part of your Apivom tenant and are not metered in the Atlas portal. The vendor's own pricing, quotas and rate limits still apply to the account you connect — Apivom paces its calls to stay inside them and shows every call in your tenant's integration log.
Where are my credentials stored and who can see the traffic?
Apivom Atlas is a stateless gateway: it never stores vendor credentials or tokens. The Apivom product that connects keeps the credential encrypted in your tenant database, and every request and response is logged in your tenant only — your data is never shared across tenants.
More Atlas integrations
AI integration guide (llms.txt)
Machine-readable markdown — feed it to your AI assistant as integration instructions.