Enable Einstein generative AI in Salesforce Setup and ensure your org has access to at least one foundation model via Model Builder.
Obtain a Salesforce OAuth access token scoped to your org; the Models API is available at the api.salesforce.com gateway and accepts standard Salesforce bearer tokens.
For single-turn text generation, POST https://api.salesforce.com/einstein/platform/v1/models/<modelName>/generations with body {"prompt": "<your prompt>", "parameters": {"maxTokens": 512, "temperature": 0.7}}.
For multi-turn chat, POST https://api.salesforce.com/einstein/platform/v1/models/<modelName>/chat-generations with body {"messages": [{"role": "system", "content": "<system prompt>"}, {"role": "user", "content": "<user message>"}]}.
Parse the JSON response; for generations the generated text is in generatedTexts[0].text, and for chat-generations it is in generations[0].content.
List available models and their capabilities by calling GET https://api.salesforce.com/einstein/platform/v1/models to discover valid <modelName> values for your org.
Known gotchas
Model names are not static across all orgs — use GET /einstein/platform/v1/models to enumerate what is available rather than hardcoding a model name.
All requests pass through the Einstein Trust Layer, which applies data masking and audit logging; outputs may be redacted if they match configured sensitive data patterns.
The Models API is rate-limited at the org level; high-volume generation workloads should implement exponential back-off on HTTP 429 responses and consider using Apex ConnectApi.EinsteinLLM methods for in-platform calls.
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp