Rosetta runs an eight-phase loop from read-only org connect through hand-off. Every phase carries decisions forward to the next; nothing is rebuilt from scratch. The whole loop runs on Cursor and is delivered by Customertimes.
The loop is the product. Every step is automated by an AI agent; every step is reviewed by a human; every decision flows into the next step via a versioned crosswalk so nothing gets lost in translation between discovery and delivery.
Read-only OAuth into the source Salesforce org. Optional connection to Event Monitoring for 90-day event-log mining. Takes minutes; no data leaves your tenant boundary unless you authorize it.
Rosetta pulls the org's metadata catalog and (optionally) anonymized event-log slices. Builds a normalized dependency graph: every Apex class → flow → field → object → layout → integration link, scored by blast radius.
Twelve specialist agents run in parallel — code quality, tech debt, automation conflict, integration architecture, security posture, governor headroom, license rightsize, process mining, dependency hotspots, custom object hygiene, flow risk, and a synthesizer that produces the Executive Brief.
Renders to a multi-document deliverable: exec brief, technical playbooks, and a JIRA-ready backlog. Reports are versioned (initial, delta1, delta7…) so subsequent scans become diffs, not duplicates.
The Discovery layer hands its findings to the Delivery layer. The Scope Builder turns the prioritized backlog into a defensible SOW — work breakdown, effort estimate, named risks, dependencies. Approve it; it becomes the contract.
The Configuration Template lands in Cursor as a structured workspace. Rosetta executes CLI-driven configuration — Apex, Flows, Permissions, Layouts, Validations — every change as a reviewable diff. The Implementation Playbook governs sequencing.
UAT Tracker auto-generates test scenarios from the Scope Builder. Defects round-trip back to Cursor as named tickets. Configuration changes apply cleanly to the sandbox, then promote to production behind a release gate.
Rosetta generates a Training Guide auto-tailored to the org's actual configuration — not a generic doc set, but a runbook for the build that just shipped. Then the Continuous Monitoring loop starts: delta scans, drift detection, weekly briefs.
Sweep, Hubbl, Swantide, Saltbox — every competitor built their own IDE-style surface. They have to keep building it, defending it, monetizing it. Rosetta took the opposite bet: ride the most rapidly-advancing AI dev environment in the industry. Cursor's roadmap becomes ours. MCP servers become ours. New models become ours the day they ship.
Salesforce, Snowflake, Jira, GitHub — as first-class MCP servers. New integrations land as new MCP servers, not new product code.
Claude, GPT, Gemini, or a private model. Switchable per task, per workspace. No model lock-in.
Every Rosetta-proposed configuration change is a reviewable patch. Approve, edit, reject — at developer-grade trust level, not click-ops blind faith.
The first scan is the cheapest way to find out whether the rest of the loop is worth running. Read-only, four hours, twelve reports.