Reason in the cloud.
The model interprets the evidence and proposes the next check or action. Each turn goes through Rapport’s gateway, with tenant model settings, session limits and usage records.
Model credentials stay server-side.THE RAPPORT AGENT WORKSPACE“Can we work
“Can we work
through this?”
A working conversation between you, cloud AI and the harness on a managed device. The model reasons through the evidence; the native agent runs checks on demand, enforces local command policy and brings held commands back for your decision.
Beta and pilot evaluation. Confirm the enabled capabilities and agent release for your environment.
CLOUD INTELLIGENCE. DEVICE CAPABILITY. HUMAN DIRECTION.“Intelligence in the cloud.
“Intelligence in the cloud.
Hands on the device.
A person in between.”
Rapport pairs cloud AI with a working harness on each enrolled device. The harness is the local software that manages the tool loop, checks proposed commands and runs permitted tools. You work with both through one conversation.
Direct the work.
Describe the problem. Challenge a hypothesis. Inspect the evidence and approve or decline held commands. Keep the investigation connected to the person experiencing the issue.
Your question. Your judgment. Your next step.Work where the issue is.
The native agent reads device state, gathers diagnostic evidence and executes permitted tools. It classifies shell commands locally before execution; the model’s request alone is not permission.
Local execution. Policy checked on the endpoint.ONE WORKING LOOP
Your question Cloud reasoning Local policy check Permitted device work Evidence for the next turn
Allowed read-only checks can run automatically. Mutating or elevated shell commands are held for approval. Commands classified as destructive or disallowed are refused.From seeing an issue to working through it.
Experience scores and fleet signals tell you where to look. The workspace lets you investigate the next question on that device, gather fresh evidence and review a tailored action. Enabled tools can address a question without a prewritten workflow for that exact issue.
Cloud reasoning uses device evidence.
Conversation context and bounded tool results travel through the gateway to the configured AI provider. Output scrubbing reduces exposure; it does not guarantee that all sensitive data stays on the device.
Explore execution and data boundariesDON’T JUST READ ABOUT THE WORKSPACEHave the conversation.
Have the conversation.
See the work behind it.
Follow four investigations adapted from internal sessions. Inspect the evidence, see where the operator changed the direction, and distinguish a verified setting from a confirmed resolution.
Agent Workspace01 / 04Look beneath the symptom
OneDrive is running. Both accounts report no upload errors and no queued work. That alone doesn’t explain why a file isn’t arriving.
I’m checking the account logs, sync folders and macOS File Provider layer together.
Inspect the supporting evidence 2 checks
shell + probeBusiness and Personal processes running; no recent app crashes; plenty of free disk space.
read_fileBoth accounts report zero upload/download errors and an empty queue.
An empty error queue is evidence to investigate, not proof that sync is working.
Files won’t sync · step 1 of 4Edited walkthroughs based on internal sessions, with identities removed. These are condensed records, not a live AI chat or a connection to your devices.
A CONVERSATION WITH SOMETHING BEHIND ITRead the explanation.
Read the explanation.
Inspect the actual work.
See the checks.
The agent can use native diagnostic checks, bounded file reads and enabled shell tools. The activity view shows tool calls and results alongside the assistant’s interpretation.
Review the exact command.
When a command is held, inspect its text and the reason for review. Approve that individual command or decline it—not a blanket plan or every later action.
Work from the result.
Read the outcome and continue the conversation. A successful command exit is not proof that the original issue is resolved; follow-up observations still matter.
UNDERSTAND THE DATA PATHLocal execution.
Local execution.
Cloud-assisted reasoning.
The endpoint runs the checks. The Rapport gateway makes the AI request. Be clear about both boundaries before enabling the workspace.
Endpoint agent
Runs the tool loop and permitted local checks.
Device-bound credentialsConversation
and tool context
and tool context
Rapport gateway
Checks endpoint, tenant and session context.
Provider credentials stay server-sideModel request
and response
and response
AI provider
Interprets observations and proposes the next step.
Review provider termsWhat crosses the boundary?
Prompts, conversation context, tool definitions and bounded tool results. These can include filenames, paths, selected file content, command output and personal information.
What needs a deployment review?
The backend stores conversation and step history. Agree data scope, retention and AI-provider handling. Pattern-based redaction reduces exposure; it does not guarantee anonymization.
THE PLATFORM AROUND THE CONVERSATIONThe workspace is where you act.
The workspace is where you act.
The platform supplies the context.
Know where to start.
Experience signals and investigations help you identify the device and the question worth asking.
Go beyond one repair.
Use the wider platform for supported remediation and staged patch work, with their own targeting and approval controls.
Keep the operating model clear.
Define collection and access boundaries, then decide how eligible agent releases reach your devices.
START WITH ONE DEVICE AND ONE REAL ISSUEBring the support question.
Bring the support question.
We’ll help scope the pilot.
Agree the device cohort, tool settings, data scope and acceptance criteria before enabling the workspace for your team.