- このガイドについて
- スタート アップ ガイド
- インストールしています。
- アップグレード
- Connecting and licensing
- 構成する
- Running workloads
- RPA processes and tests
- Agents, functions, and API workflows
- 上限量
- Securing and governing
- Robot のポリシーの設定
- Tracing and data controls
- CrowdStrike との連携
- 監視
- トラブルシューティング
- 参照
Tracing and data controls
What data UiPath Robot 2026.10 sends from your machine to the UiPath Platform, what traces capture, and the controls that limit it.
Running a workload on your own machine changes where the work happens. It doesn't, by itself, change what the job reports: status, logs, and traces still reach the platform.
The data your automation reads and writes stays where the automation puts it, unless the job logs it, traces it, or sends it to a model. Those three paths, plus the Robot's own telemetry, are where data leaves the machine.
What leaves the machine
| データ | Sent to | Sent for | 次の値を含む |
|---|---|---|---|
| Job status and logs | Orchestrator | Every job | What the Robot and your automation write to the log |
| トレース | The UiPath Platform, shown in Orchestrator and in agent traces | Agent, function, and API workflow jobs started from Orchestrator | Inputs and outputs of each step, and the prompts and responses of each model call |
| Activity traces | The UiPath Platform | RPA processes and test cases in Windows and cross-platform projects, unless the UiPath Platform's trace service turns activity tracing off | The activities the job ran |
| Model calls through the LLM Gateway | The AI Trust Layer, then the model provider | Agents and activities that use UiPath-provided models | Prompts and responses |
| テレメトリ | UiPath | Every robot, unless disabled | Usage data, such as activity usage and workflow and library names |
UiPath treats traces as customer data. They are kept for 90 days unless your AI Trust Layer policy sets a shorter time to live.
What traces capture
Traces are the most detailed of the three paths. By default, a trace records:
- The prompts and responses of every model call an agent makes, whichever model it calls. That includes your own model and a model running on your premises.
- The inputs and outputs of each traced step in a coded agent or function, and of most SDK calls.
- Calls to the LLM Gateway made through the SDK, including the prompt.
The SDK redacts a narrow set of values on its own: asset values, credentials, and secrets, the output of connection token calls, and attachment content. Anything else your code passes to a traced step is recorded as is.
A coded agent or function sends traces to UiPath only when it runs as a job from Orchestrator, or during an evaluation that reports to Studio Web. A local run with uipath run sends none.
Controls for coded agents and functions
Each control limits a different part of the trace. Only the first one stops traces from leaving the machine.
| 制御 | 効果 | スコープ |
|---|---|---|
UIPATH_TRACING_ENABLED=false | Stops sending traces to UiPath. Spans are still created on the machine. | Every span from the job. Requires uipath 2.8.36 or later. Any value other than true, including 1, turns sending off. |
@traced(hide_input=True, hide_output=True) | Replaces that step's input or output with a redaction placeholder | That step only. A value returned into agent state can still appear in the spans around it. |
OPENINFERENCE_HIDE_INPUTS=true, OPENINFERENCE_HIDE_OUTPUTS=true | Redact inputs or outputs on model and framework spans | Framework spans only, such as LangChain. Doesn't affect @traced steps or SDK calls. Accepts true or false. |
OPENINFERENCE_HIDE_INPUT_MESSAGES=true, OPENINFERENCE_HIDE_OUTPUT_MESSAGES=true | Remove the message lists from model spans | The prompt and response text remain in the span's input and output. Not enough on its own. |
The environment variables take effect only when they are set in the job's environment before the job starts. On a self-managed robot, set them in the Environment Configuration field of the process in Orchestrator, or of the job when you start it, one NAME=value pair per line. The Robot passes them to the job. For details, refer to Managing processes and Managing jobs.
For low-code agents, the AI Trust Layer policy controls how long traces are kept and how much trace detail reaches Insights. For those settings, refer to Settings for AI Trust Layer policies.
Activity tracing for RPA processes
RPA processes and test cases record activity traces by default. The UiPath Platform's trace service controls whether the Robot sends them and for which jobs. If the Robot can't get that setting, it keeps activity tracing on. Windows - Legacy projects don't record activity traces.
To turn activity tracing off on a machine, set the UIPATH_DISABLE_ACTIVITY_TRACING system environment variable to True, then restart the Robot Service.
テレメトリ
The Robot sends usage telemetry to UiPath to keep the service running and to improve it. The UiPath Platform Installer applies its own telemetry setting to the Robot: when you install or change the Robot with TELEMETRY_ENABLED=0, the installer sets Telemetry.Enabled to false in the Robot's UiPath.config file. The installer applies the same setting again when it updates the Robot, so a value you set by hand in that file doesn't survive an update. Set telemetry through the installer parameter instead. You can also set Telemetry.Enabled in that file yourself, as described in Configuration files. To turn telemetry off for the machine regardless of that file, set the UIPATH_TELEMETRY_DISABLED system environment variable to true. It takes precedence over Telemetry.Enabled. For details, refer to Use of customer data with Robot.