Can You Use Claude or ChatGPT in Salesforce Without Agentforce? 4 Alternatives
Yes. Four Agentforce alternatives: Prompt Builder with your own LLM, ChatGPT or Claude through Salesforce Hosted MCP Servers, Apex callouts through Named Credentials, or a Salesforce-native AppExchange app such as GPTfy. They differ on agents, masking, Data 360 and pricing.
Last updated:
Can you use Claude or ChatGPT in Salesforce without Agentforce?
Yes. There are four ways to put Anthropic Claude, OpenAI GPT or Google Gemini to work on Salesforce data without building agents in Agentforce:
- Prompt Builder with your own LLM: Salesforce's prompt templates, running on a model you add in AI Models.
- ChatGPT or Claude through Salesforce Hosted MCP Servers: your users stay in ChatGPT or Claude, which read and update Salesforce on their behalf.
- Your own build: Apex callouts to the model through Named Credentials, or MuleSoft's AI connectors.
- An AppExchange app such as GPTfy, which runs inside your org and connects to your own model.
GPTfy is a Salesforce-native Agentforce alternative on AppExchange: bring your own model (BYOM), no Data 360, fixed per-user pricing.
Disclosure: GPTfy publishes this page and sells one of these options. Each route is described from its vendor's own documentation, linked under Sources.
How do the four routes compare?
| Prompt Builder + your LLM | ChatGPT or Claude via Hosted MCP | Your own Apex build | GPTfy (AppExchange) | |
|---|---|---|---|---|
| Runs agents? | No, prompt templates. Agents come from Agentforce | The agent runs in ChatGPT or Claude | Whatever you build | Yes, inside Salesforce |
| Models | Your model from Amazon Bedrock, Azure OpenAI, OpenAI or Vertex AI, or any endpoint via the LLM Open Connector | The client's models: OpenAI in ChatGPT, Anthropic in Claude | Any model your provider offers | OpenAI, Azure OpenAI, Claude, Gemini, Bedrock, Mistral, Llama, DeepSeek and more |
| Where data is processed | Your provider, routed through the Models API and Trust Layer | The AI vendor's service | Wherever you send it | Your provider. Raw data stays in Salesforce |
| Masking before the model | Yes for templates run directly; disabled when an agent calls them | None documented for this route | Only what you build | Yes, four layers: field, regex, blocklist, Apex |
| Data 360 required? | Yes, provisioned and enabled | No | No | No |
| Pricing | Per prompt, in Flex Credits, plus your provider's token charges | AI client seats, plus Salesforce usage (intended for Flex Credits customers) | Provider tokens plus developer time | $20 to $50 per user per month, 100-user minimum, plus provider tokens |
| Effort | Admin setup | Under 30 minutes, per Salesforce | A development project | Install from AppExchange, add a Named Credential, configure prompts and masking |
Prompt Builder with your own LLM
In AI Models (formerly Einstein Studio) you can add a model from Amazon Bedrock, Azure OpenAI, OpenAI or Vertex AI, or any endpoint through the LLM Open Connector, then use it in Prompt Builder templates. Requests still route through the Models API and the Einstein Trust Layer, and masking applies when a template runs directly or through Flow.
Limits to know: Prompt Builder runs templates, not agents. When an Agentforce agent calls a template, Salesforce documents that masking is disabled for agents. Setup follows the Einstein generative AI path, so Data 360 must be provisioned and enabled, and usage is metered per prompt in Flex Credits. See Does Agentforce require Data Cloud?
ChatGPT or Claude through Salesforce Hosted MCP Servers
Salesforce Hosted MCP Servers let a compatible client, such as ChatGPT or Claude, act in a Salesforce org on behalf of each authorised user through per-user OAuth. Salesforce announced general availability in April 2026 for Enterprise Edition and above, and says setup takes less than 30 minutes.
Limits to know: the records the client reads are processed in the AI vendor's service, and none of Salesforce, OpenAI or Anthropic documents masking before the model on this route. Salesforce says Hosted MCP Servers are intended for customers with Flex Credits, so there are two bills: AI client seats and Salesforce usage.
Your own build with Apex and Named Credentials
Developers can call any model API from Apex through a Named Credential, or use MuleSoft's Inference or AI Chain connectors. You get any model and no Data 360 dependency. You also own everything else: prompt management, masking, audit logs, retries and model changes. This route suits teams with Salesforce developers and one or two narrow, high-volume use cases.
An AppExchange app such as GPTfy
GPTfy installs as a managed package. You connect your own model through a Named Credential, on your own OpenAI, Anthropic, Azure, AWS or Google Cloud contract, and can route different prompts to different models. Data is masked in four layers before the callout, every AI interaction is logged in Salesforce, and Enterprise and above include AI agents that read and update records and call Flow and Apex. GPTfy does not require Agentforce or Data 360.
Other AppExchange apps take this route too, with different trade-offs: Ortoo Orchestrator, Peeklogic AI Orchestrator and Oliva.AI by LG CNS (OpenAI only). They are compared on Agentforce alternatives.
Setup detail: Bring your own model in Salesforce. Pricing: from $20 per user per month with a 100-user minimum, model usage billed by your provider. See GPTfy pricing.
Which route should you choose?
Unlike Agentforce competitors that replace Salesforce, all four routes keep your data and users on Salesforce. Pick by the job:
- Single-step tasks inside Salesforce's own tooling (summaries, drafts), with Data 360 already enabled: Prompt Builder with your own LLM.
- Users who already live in ChatGPT or Claude and need Salesforce as a tool there, with no masking requirement: Hosted MCP Servers.
- One narrow, high-volume use case and in-house developers: your own Apex build.
- Agents inside Salesforce on your own model, masking before the model, no Data 360, per-user pricing: an AppExchange app such as GPTfy.
Sources
- Salesforce Help: add a foundation model (BYO LLM)
- Salesforce Help: Prompt Builder
- Salesforce Help: data masking for agents
- Salesforce Developers: Models API data masking
- Salesforce Help: generative AI usage and billing
- Salesforce Developers: Hosted MCP Servers overview
- Salesforce Developers blog: Hosted MCP Servers GA
- Claude Help: set up Salesforce in Claude
- Apex Developer Guide: Named Credentials
- MuleSoft Inference Connector
- GPTfy on AppExchange
Key takeaways
Yes, without building in Agentforce
Claude, GPT and Gemini can all work on Salesforce data through Prompt Builder, Hosted MCP Servers, your own Apex, or an AppExchange app.
Only some routes run agents
Prompt Builder runs prompt templates, not agents. Hosted MCP Servers put the agent in ChatGPT or Claude. Your own build and AppExchange apps such as GPTfy can run agents inside Salesforce.
Masking differs by route
Trust Layer masking applies to prompt templates run directly. No masking is documented for the Hosted MCP route. GPTfy masks in four layers before the model call.
Data 360 differs by route
Prompt Builder with your own LLM follows the Einstein generative AI setup, which needs Data 360 provisioned and enabled. Apex callouts and GPTfy do not.
FAQ
Yes. GPTfy is a Salesforce-native Agentforce alternative on AppExchange. It runs AI agents inside Salesforce on a model you bring (OpenAI, Anthropic Claude, Google Gemini and others), masks data before the model call, does not require Agentforce or Data 360, and is priced per user: $20, $30 or $50 per user per month with a 100-user minimum, plus your model provider's charges.
Yes. You can add Claude through Amazon Bedrock as your own model in Prompt Builder, connect the Claude app to Salesforce through Salesforce Hosted MCP Servers, call the Anthropic API from Apex through a Named Credential, or use an AppExchange app such as GPTfy that connects to Claude on your own contract and runs agents inside Salesforce.
Yes. Add an OpenAI or Azure OpenAI model in Prompt Builder, connect ChatGPT through Salesforce Hosted MCP Servers, call the OpenAI API from Apex, or use an AppExchange app such as GPTfy with your own OpenAI or Azure OpenAI key.
With Salesforce's own BYO LLM in Prompt Builder, yes: setup follows the Einstein generative AI path, so Data 360 (formerly Data Cloud) must be provisioned and enabled. Apex callouts through Named Credentials and GPTfy do not require Data 360.
Prompt Builder masks data for templates run directly, per your Trust Layer settings, but Salesforce documents masking as disabled for agents. No masking is documented for the Hosted MCP route. GPTfy masks in four layers (field, regex, blocklist and Apex) before every model call. With your own Apex build, masking is whatever you implement.
With Prompt Builder you can add several models in AI Models. With GPTfy you can connect several providers and route different prompts to different models without rebuilding them. With a custom Apex build, switching means changing your code.
See your own model working in Salesforce
Connect Claude, GPT or Gemini on your own contract and run it on your Salesforce data, with masking.
Explore More
Bring Your Own Model
Connect OpenAI, Claude, Gemini and more through Named Credentials.
Which LLM Does Agentforce Use?
Agentforce's default models and the org-level options.
Agentforce Alternatives
Every route to AI in Salesforce, compared on the same seven criteria.
Does Agentforce Require Data Cloud?
Provisioned versus implemented Data 360, with Salesforce sources.