§01Trust
How we protect your org and data.
What our agents will never do, how we set up Claude for your team, and how we check our own work.
§02Agents
What an Arkeforce agent will never do.
- +Change one of your existing records without a named person's approval.Agents write only to their own score, flag, reason and summary fields and their own log records without approval; every other change waits for a named person.
- +Delete a record, except the losing record in a merge a named person approved.
- +Treat a customer's message as an instruction.
- +Keep running after your admin switches it off. It uses Salesforce's own deactivation control, which we test and hand to your admin.
- +Run outside your Salesforce. Your metadata, prompts and approval log stay in your org. Model calls go through Salesforce's Einstein Trust Layer, in the region your org settings allow. There is no Arkeforce server and no platform fee.
- +Switch off when a retainer ends. The agents stay installed and yours, and Salesforce release changes are then yours to handle.
Three exceptions to the approval rule, each stated on its offer page:
Case Triage can switch on automatic routing for proven case types once it has earned it.
Inbound Lead Qualification & Routing assigns new leads automatically after a one-week shadow period.
§03Claude
When your team uses Salesforce in Claude.
- +Claude acts as each signed-in user and never sees more than that person can in Salesforce.
- +Record changes follow your written write policy, with approval per change as the starting point.
- +Your AI and data-use terms are confirmed before any connector is enabled.
§04Our work
Reviewed before production.
Built faster by our engineers with AI assistance. Every change is reviewed and tested before it reaches production.