🔗Telemetry

Boxxy Terminal includes a =privacy-first=, open-source telemetry system designed to help us improve the application's performance, stability, and especially the AI features, such as tracking tool acceptance and rejection rates. Basically, telemetry is your way of contributing to the project!

We believe in =complete transparency= regarding the data we collect, so we publicly share this data in REAL TIME with the hope that it will prove valuable for the open-source community!


🔗Privacy Principles

  • Opt-In/Opt-Out: Telemetry is strictly configurable. While it may be enabled by default during preview phases to gather initial feedback, you can easily disable it at any time via =Preferences=.
  • Zero PII (Personally Identifiable Information): We never collect file paths, file contents, usernames, IP addresses, prompt contents, or terminal outputs.
  • Anonymous Identifiers:
    • install_id: A random, anonymous identifier generated once when you first run the app. This helps us count unique installations without knowing who you are.
    • session_id: A random identifier generated each time you launch the app, kept only in memory for that session. This helps us understand session lengths and distinguish between many users vs. a single user launching the app multiple times.
  • Open Standard: We use the industry-standard =OpenTelemetry (OTel)= protocol, ensuring our data collection methods are open and auditable.

🔗What Data We Collect

We focus on collecting high-impact, low-cardinality metrics to provide value while keeping the data footprint extremely small.

Here is the exhaustive list of the metrics we track when telemetry is enabled:

CategoryMetric NameDescriptionData Attributes Collected
Growth & Usageapp.launchTriggered once per application startup.os, arch, pkg_type (e.g., flatpak, native), version, shell (e.g., bash, zsh, fish)
AI Usageai.tokensThe number of tokens consumed during an AI interaction.model_name, role (input/output), feature
AI Performanceai.invocationsA count of AI requests initiated by the user.model_provider, model_name, feature
AI Qualityai.latencyThe time it takes (in milliseconds) to receive the first token back from the AI model.model_name, provider, feature
Product Featurestool.useA count of which AI tools are being utilized by the agent.tool_name (e.g., sys_shell_exec, file_read)
Retentionclaw.session_resumeTriggered when a previous BoxxyClaw session is restored.session_type (e.g., normal, pinned)

🔗How the Data is Used

The data collected is securely transmitted to our backend infrastructure. We use this aggregated, anonymous data to:

  • Understand which features and AI models are most popular.
  • Diagnose performance bottlenecks (e.g., high latency with specific providers).
  • Prioritize development efforts for different operating systems and deployment methods.
  • Publish =public, aggregated community statistics=.

Note: Gathering information and policies might change, so please check back, or just check the code! 😊