Telemetry

Magia has three separate telemetry systems: a local OpenTelemetry collector, Sentry crash reporting, and PostHog analytics. All three are opt-in for self-builds: if the relevant environment variable is not set at compile time, that system is entirely disabled at runtime — no data leaves the machine.

---

Telemetry Components at a Glance

Component	Purpose	Activated by	Default (official builds)	Default (self-builds)
OpenTelemetry (OTel)	Session metrics, token usage, tool calls	MAGIA_OTEL_ENDPOINT build-time env var	Enabled (collector runs locally)	Collector runs locally; data stays local — no export endpoint
Sentry	Crash reports and error diagnostics	MAGIA_SENTRY_DSN build-time env var	Enabled	Disabled (no DSN compiled in)
PostHog	Anonymous product analytics	MAGIA_POSTHOG_KEY build-time env var	Enabled	Disabled (no key compiled in)

Self-build guarantee

If you build Magia from source without setting the environment variables above, zero telemetry data is sent to any external server. The OTel collector still runs locally (so the in-app metrics dashboard works), but it has no export destination.

---

OpenTelemetry

Architecture

Magia embeds a lightweight OTLP collector inside the Tauri backend. Agent processes (Claude CLI, Gemini CLI, etc.) are started with OTEL_EXPORTER_OTLP_ENDPOINT pointing to this collector. The collector receives metrics and log records from agents and processes them in-process.

Two transport protocols are supported:

Protocol	Description	Used by
HTTP (OTLP/HTTP)	REST over TCP; default port 4318	Agent processes via OTEL_EXPORTER_OTLP_ENDPOINT env var
gRPC (OTLP/gRPC)	Protobuf over a Unix socket	Agent processes on platforms that support Unix sockets

The port can be configured with otelCollectorPort (default 4318). If that port is unavailable, the server tries the next available port automatically.

What Data is Collected

Metrics

Metric	Description
input_tokens	Number of input tokens consumed by the model in a session.
output_tokens	Number of output tokens generated.
cache_read_tokens	Number of tokens served from the prompt cache.
cache_creation_tokens	Number of tokens written to the prompt cache.
total_cost_usd	Estimated USD cost of the session (computed from token counts and model pricing).
context_window_used	Fraction of the model’s context window consumed (0.0–1.0).
context_window_size	Total context window capacity in tokens for the active model.

Logs / Events

Event	Description	Controlled by
Tool call start / end	Tool name, start time, duration, and status (success, error, pending).	Always collected when OTel is enabled.
User prompts	The text of user messages sent to the agent.	otelLogUserPrompts (default: off).
Tool call details	Arguments and outputs of tool calls.	otelLogToolDetails (default: off).

Prompt and tool data

otelLogUserPrompts and otelLogToolDetails are disabled by default because prompts and tool arguments may contain sensitive data (file contents, credentials, etc.). Enable them only on private infrastructure you control.

Configuration Settings

Setting	Key	Default	Description
Enable collector	otelEnabled	true	Start the local OTLP collector on app launch. Disabling this also removes the export endpoint from agent processes, so no metrics are collected at all.
Collector port	otelCollectorPort	4318	TCP port for the HTTP OTLP collector.
Metric export interval	otelMetricExportIntervalMs	5000	How often (ms) metric batches are flushed.
Logs export interval	otelLogsExportIntervalMs	2000	How often (ms) log record batches are flushed.
Log user prompts	otelLogUserPrompts	false	Include prompt text in log records.
Log tool details	otelLogToolDetails	false	Include tool arguments and results in log records.

Disabling OTel

Toggle Settings → Telemetry → Enable OpenTelemetry off, or set otelEnabled: false in settings.json. Agent processes will not receive an OTEL_EXPORTER_OTLP_ENDPOINT variable and will not attempt to export telemetry.

---

Sentry (Crash Reporting)

Sentry captures unhandled panics and JavaScript errors. Reports include a stack trace, the OS version, the Magia version, and any Sentry breadcrumbs set before the crash. No user-generated content (prompts, code, file paths) is included in crash reports.

Sentry is activated at compile time by the MAGIA_SENTRY_DSN environment variable. If this variable is empty or absent, the Sentry SDK is never initialized — not even in a disabled state.

Disabling Sentry

• Self-builds: No action needed. Sentry is not compiled in.
• Official builds: Toggle Settings → Telemetry → Crash Reporting off. This sets crashReportingEnabled: false and the SDK will not send any reports.

---

PostHog (Analytics)

PostHog collects anonymous usage events such as:

• App launch and session counts
• Feature adoption signals (e.g. which providers are used)
• Funnel events (e.g. onboarding completion)

Events do not include content from sessions, prompts, file contents, or any personally identifiable information. A random anonymous identifier is generated per installation.

PostHog is activated at compile time by the MAGIA_POSTHOG_KEY environment variable. If this variable is empty or absent, the PostHog client is never initialized.

Disabling PostHog

• Self-builds: No action needed. PostHog is not compiled in.
• Official builds: Toggle Settings → Telemetry → Analytics off. This sets analyticsEnabled: false and the client will not send any events.

---

Summary: How to Fully Opt Out

Scenario	What to do
Self-build (source)	Nothing — all remote telemetry is disabled by default.
Official build — disable everything	Turn off Crash Reporting, Analytics, and OpenTelemetry in Settings → Telemetry.
Official build — keep local metrics only	Disable Crash Reporting and Analytics; keep OpenTelemetry enabled. Data stays on-device.
Custom OTel export to your own backend	Set MAGIA_OTEL_ENDPOINT at build time to your collector URL and keep otelEnabled: true.