The workspace agent is Home: you land in a conversation, not a dashboard. Ask about your organization in plain language — “What changed in the org last week?”, “Who can edit Opportunity Amount?”, “Which fields on Case are never used?” — and it reasons over your synced Salesforce catalog and your documentation corpus, running tools in a loop until it has a grounded answer. Starter chips on the empty state (Field impact, Audit this org, Recent changes, Stale docs, Coverage gaps) give you one-click first questions.
Conversations are persisted, private threads — the thread list on the left groups them by day, you can leave and come back, and other members can’t see your threads (“Private to you · org-scoped”). Each thread’s menu offers Rename, Capture as skill (turn a good investigation into a reusable draft skill under Agents → Skills), and Delete.
In the composer: type @ to mention a Salesforce object, field, or flow and pin it into context; type / to pick a playbook; attach screenshots or text files; dictate if your browser supports it; and pick the AI model for the chat from the models your org enabled in Settings → Models. Enter sends, Shift+Enter adds a line.
The workspace agent is read-only. It never modifies your Salesforce org or your docs. Its single write path is Save as doc, which goes through the normal change request review flow. Secrets are never echoed. If your synced source or records contain credentials — API keys, tokens, passwords, client secrets — the agent reports where they live and flags them as a security risk, but never prints the full value in its answer or inside a code snippet. It treats your documentation as leads to verify, not proof: for questions about how your org is configured today, it confirms against your live org and metadata before asserting.
What it can see
The agent chooses from some twenty read-only tools. As it works, tool activity streams live into the transcript — collapsed by default under a summary like “Thought & used 4 tools · 12s”, expandable down to each tool’s input and output — so you can see exactly what it looked at. A Usage chip under each answer breaks out the model, tokens, credits, and cost of that turn.
Source search
Impact analysis
Change history
Security and access
Field hygiene and org health
Integrations and docs
Live tools (audit_trail, security_health_check, field_utilization) call Salesforce using your organization’s connection. Field utilization returns counts only — the agent never reads record contents. See Salesforce integration. Slash playbooks
Playbooks are pre-built investigations you launch with a slash command:
Prompt commands
Shape how the agent responds within a thread:
Charts and dashboards
Ask for a chart or a dashboard in plain language — “chart Opportunities won by month”, “build a pipeline dashboard by stage and owner” — and the agent renders it inline in the chat, in your workspace theme. The charts it creates also appear in Analytics, where you can open and filter them any time. Like everything else the agent does over your records, this is read-only — it never changes your Salesforce data.
Charts require an analytics source to be connected first — see Turn analytics on. Until then the agent will point you to Connections instead of rendering. Save an answer as a doc
When the agent produces something worth keeping — an audit summary, an integration inventory, an architecture explanation — click Save as doc. The answer is captured as a change request, filed under a topic folder inside workspace-notes/ in your docs repo. Like everything else, it ships only after review.
Run /audit weekly and save the result — you get a reviewed, timestamped history of your org’s health in your own Git repository.
