Telemetry pricing punishes you for capturing the thing that explains the failure. Watchpoint charges by fleet size, so the incentive to instrument deeply stays intact.
The full forensics core, running entirely on your infrastructure.
For fleets past the point where one engineer can triage by hand.
Minimum 10 robots. Robots in a bench or CI fleet aren't counted.
Talk to usFor teams with a safety case, an auditor, or an air gap.
We deliberately built self-hosted first, because most robotics teams cannot export customer camera data. A managed option is planned for teams without that constraint. It isn't built yet, and we'd rather say so than take a waitlist deposit on it.
| Capability | Community | Team | Enterprise |
|---|---|---|---|
| Self-hosted deployment | |||
| Model collector + ROS 2 + edge agent | |||
| System rules engine (7 rules) | |||
| Shipped AI rules (AI-001–003) | |||
| Replay bundle export | |||
| Full AI taxonomy (AI-004–008)roadmap | |||
| Replay sandboxroadmap | |||
| Cross-fleet baselines | |||
| Priority support SLA | |||
| SSO / SAML + audit log | |||
| Air-gapped deployment support | |||
| Custom rules built for you |
Items marked roadmap are specified but not merged. Team and Enterprise include them as they ship; we don't bill for them before that.
A physical unit running a collector and reporting under its own device ID. Bench rigs, simulation, and CI fleets are free and uncounted — we don't want to price you out of testing.
No. It's the complete forensics core: every collector, the system rules engine, the AI rules that have shipped, and bundle export. Paid tiers add fleet-scale analysis, the replay sandbox, and support — not access to your own data.
The core is Apache 2.0 — every collector, the API, the rules engine, and the dashboard. Anything that runs on your robot is Apache-licensed without exception, so you can audit it before you deploy it. Commercial features like SSO and cross-fleet rollups are licensed separately and live outside that tree.
No. Self-hosted means the stack runs in your VPC with your Postgres and your storage. Model weights are hashed for lineage, never uploaded. With no Anthropic key set, the LLM summary degrades to deterministic rules text and nothing leaves your network at all.
You fall back to Community. It's self-hosted, so your data stays where it already is — there is no hostage situation and no export scramble.
Volume pricing makes you capture less exactly when you should capture more. The whole product depends on the failure frame being in the buffer, so we refuse to charge in a way that discourages that.