Telemetry Disclosure

Effective: March 17, 2026

1. Why We Collect Telemetry

TitanOcta is local-first AI infrastructure. Telemetry submission is a condition of use across all tiers — free and paid. The data feeds community Grafana dashboards and the benchmark layer so every user benefits from aggregate performance insights across real hardware.

2. What We Collect

Here is the complete list. If it is not on this table, we do not collect it.

Data Point	Description	Collected
Hardware specs	CPU model, GPU model/VRAM, total RAM, disk type	Yes
Model identifier	Which LLM was routed to (e.g. llama-3-70b, claude-sonnet)	Yes
Performance metrics	Tokens/sec, latency, throughput, VRAM usage during inference	Yes
Benchmark data	Standardized benchmark scores when running evals	Yes
Deployment stats	Agent start/stop events, uptime, restart counts	Yes
Crash reports	Stack traces, error codes, OS version at time of crash	Yes
Prompts	Your messages to agents	Never
Session content	Agent responses, conversation history, outputs	Never
API keys	Your third-party provider credentials	Never
Agent memory	Persistent memory stored by your agents	Never
File contents	Documents, code, or data processed by agents	Never

The line is simple: We collect how your hardware performs running models. We never collect what you say to those models, what they say back, or any credentials you use to access them.

3. How Telemetry Is Transmitted

Telemetry is sent over TLS to our ingestion endpoint.
Data is tagged with an anonymous installation ID — not your account email or name.
Payloads contain only the fields listed above. No ambient data collection.
You can inspect the exact payload in your agent logs before it leaves your machine.

4. How We Use It

Community benchmarks: Aggregated, anonymized stats published on Grafana dashboards available to all users.
Routing optimization: Flow uses aggregate performance data to improve model routing recommendations.
Stability: Crash reports help us identify and fix bugs faster.

We do not use telemetry for advertising, user profiling, or model training.

5. What We Do Not Do

We do not sell telemetry data to third parties.
We do not correlate telemetry with your identity for ad targeting.
We do not train AI models on any user data — telemetry or otherwise.
We do not share individual hardware profiles externally. Only aggregates.

6. Retention

Raw telemetry is retained for up to 90 days, then aggregated and anonymized. Aggregated benchmark data is retained indefinitely as part of the community dataset.

7. Your Controls

Inspect: All telemetry payloads are logged locally. You can review exactly what is sent.
Non-essential telemetry: Crash reports and deployment stats can be disabled in account settings.
Core telemetry: Hardware specs and model performance data are required as a condition of use per the Terms of Service.

8. Changes

If we ever expand what telemetry we collect, we will update this page and notify all users at least 14 days before the change takes effect. The boundary — no prompts, no session content, no keys, no memory — does not change.

9. Contact

Questions about telemetry? Email legal@synoct.com.

Business and project inquiries: kamal@synoct.com