# Claude models, grounded in your Salesforce org. > SyncOnAI 360 can run its chat and specialist agents on Anthropic's Claude models. The chat and the agents each have their own model setting, paid plans include AI with the model chosen per task, and the Free plan uses your own Anthropic key. Anthropic does not train on API traffic, whichever account a request uses. Source: https://synconai360.com/connectors/anthropic ## Key facts - **2**: Separate model settings: chat and agents - **0**: Training on API traffic - **Free**: Plan runs on your own Anthropic key - **KMS**: Encryption for any key you store ## Why a strong model alone is not enough for Salesforce Claude is very good at Salesforce in general. What it lacks on its own is your org: the objects, the automation, the conventions and the history. And on its own it has no route to a safe deploy. - **No org context.** A model answering from general knowledge describes a Salesforce org, not yours. - **No guardrails.** A model that can write metadata has no checks, approval or rollback by itself. - **Key sprawl.** Each person pastes their own API key into their own tools, with no workspace control. - **Unclear data use.** Teams need to know whether their org details train someone's model. ## How do Claude models work inside SyncOnAI 360? Every request carries a compact profile of the connected org, and the model reads the synced metadata and source through tools: searching components, looking up dependencies, reading health findings. Each answer shows the tools it used. Anything the model drafts, a Flow, a validation rule, an Apex class, becomes a proposal with named checks, validation against the org and approval for production. The model writes; the platform decides what reaches the org. - Org profile in every request - Tools that read the synced org - Drafts become proposals ## How do you choose models for the chat and the agents? Under Settings, Models, the workspace chooses the model the chat uses and, separately, the model the agents use, from Anthropic and OpenAI. Chat and agents can run on different models. When AI is included in the plan, the right model is picked per task: a light model for quick answers, a stronger one for building and the strongest for hard reviews. - Chat and agents set separately - Anthropic or OpenAI - Per-task choice when included ## How does bring your own Anthropic key work? On the Free plan, AI runs on your own Anthropic or OpenAI key, billed by that provider. Paste the key under Settings, API keys; one key per workspace, and Test checks that it works. Keys are encrypted under a key held in a cloud key management service and are never shown in full after saving. Paid plans include AI, so no key is needed there. - One key per workspace - Test before you rely on it - Encrypted, never shown in full ## Does Anthropic train on your Salesforce data? No. Anthropic does not train on API traffic, whichever account a request uses. SyncOnAI 360 sends the model the org context a request needs, metadata and source, and never stores your CRM records. Anthropic is listed on the subprocessors page with what it receives and where, alongside every other company that processes data for SyncOnAI 360. - No training on API traffic - Metadata, not CRM records - Listed as a subprocessor ## Can you use Claude outside SyncOnAI 360 too? Yes. Claude Desktop and other Claude clients can connect to your orgs through the read-only Model Context Protocol server, with tokens scoped per org and every call logged. Inside the workspace Claude can draft changes as proposals; outside it, Claude reads. Both work from the same synced copy of the org. - Claude Desktop through MCP - Read-only outside the workspace - One synced org behind both ## How do Claude models follow your org's conventions? Org rules set naming conventions, forbidden patterns and the preferred automation type for each org, and anything the model drafts is built inside them. Before building, the chat checks what already exists, so it extends what is there rather than adding a second way of doing the same thing. Specialist agents add their own instructions and skills, and their tools can be allowed, denied or held for confirmation, whichever model runs them. - Org rules the model must follow - Existing solutions checked first - Agent tools under your control ## Using Claude models 1. **Choose.** Set the chat and agent models under Settings. 2. **Connect a key.** On Free, add your Anthropic key; paid plans include AI. 3. **Ask.** Answers come from the synced org, with the tools shown. 4. **Propose.** Changes become proposals with checks and approval. ## Claude on Salesforce, on its own versus in SyncOnAI 360 | Without | With SyncOnAI 360 | |---|---| | General Salesforce knowledge | Your org's metadata and source in every answer | | Output pasted into Setup | Drafts become checked, approved proposals | | A key in every person's tools | One workspace key, encrypted, or AI included | | Uncertain data use | No training on API traffic, records never stored | ## Who chooses Claude models - **Salesforce developers.** Use Claude for Apex, components and reviews, grounded in the org's own patterns. - **Salesforce admins.** Ask about the org and draft configuration in plain language, with every change held as a proposal. - **Independent consultants.** Start free on your own Anthropic key, then move to a plan with AI included. - **IT and platform leaders.** Standardise on one governed way to use Claude on Salesforce. ## Frequently asked questions ### Does SyncOnAI 360 use Claude? It can. The chat and the agents can each run on Anthropic's Claude models, chosen under Settings, Models. ### Do we need an Anthropic key? Only on the Free plan. Paid plans include AI. ### Does Anthropic train on our data? No. Anthropic does not train on API traffic. ### Can chat and agents use different models? Yes. Each has its own model setting. ### How is our key stored? Encrypted under a key held in a cloud key management service, and never shown in full after saving. ### Can we use Claude Desktop with our org? Yes, read-only, through the Model Context Protocol server. ### Does the model see our CRM records? SyncOnAI 360 keeps metadata, not CRM records. Record data is only read live when a question needs it. ### Can the model deploy changes? No. Anything it drafts becomes a proposal with checks, validation and approval. ### Which plans include AI? Pro, Team and Business include AI, so no key is needed. Free runs on your own Anthropic or OpenAI key, and Enterprise can include AI or use your own provider agreement. ### Can we switch provider later? Yes. Change the model setting at any time; your orgs, history and proposals stay as they are.