Effective July 9, 2026
This policy explains how the RoBrain Claude Code plugin and the open-source, self-hosted RoBrain tooling (together, the “Software”) handle your data. The Software is published by Rory Plans, Inc. (“Rory Plans,” “we”) and licensed under Apache 2.0.
RoBrain is self-hosted. Your session content and the decisions extracted from it are stored in your own database and never sent to Rory Plans. The only outbound network calls are to the third-party LLM and embedding providers you configure with your own API keys.
When installed in a coding tool (Claude Code, Cursor, GitHub Copilot, or Codex CLI), RoBrain passively captures your session turns and extracts decisions from them — for example, an architectural choice and the alternatives your team rejected. Extraction produces structured records (decision text, rationale, rejected alternatives, affected file paths, timestamps, and confidence scores).
The Software does not capture your source code or file contents, secrets or credentials (these are actively redacted at capture and again at ingest), or anything outside your sessions with the agent.
Extracted decisions are stored in a PostgreSQL database that you host and control — by
default a Docker container on your own machine or infrastructure. Rory Plans does not receive, store, or have
access to this data. Access to the local store is guarded by an API key (PERCEPTION_API_KEY) that
the Software generates on your machine.
When you use the self-hosted Software, Rory Plans collects nothing. The plugin and CLI communicate only with the local Perception service and the third-party providers listed below. We do not operate any telemetry, analytics, or phone-home in the self-hosted Software.
To extract and index decisions, the Software calls the model and embedding providers you configure using your own API keys. Your session content is sent to these providers solely to perform extraction and generate embeddings, and is subject to their privacy policies:
You choose which providers to use. If you run RoBrain in fully-local mode — with the LLM and embeddings served by a local runtime such as Ollama, LM Studio, or vLLM — no session content leaves your machine at all.
Extracted decisions are used only to provide RoBrain’s features: injecting relevant prior decisions at session start, warning you before an agent repeats a previously rejected approach, and running corpus-wide judgment (contradiction and drift scans). The data is not used to train any model.
Because the data lives in your own database, you control retention. Decisions persist until
you remove them; you can inspect, supersede, or delete records with the RoBrain CLI (e.g.
robrain review) or by operating on your PostgreSQL instance directly. Uninstalling the plugin
stops further capture; existing records remain in your database until you delete them.
Requests to the local Perception service require the PERCEPTION_API_KEY bearer token, and
secrets are scrubbed from content before it is embedded or stored. See the project
security notes for details.
This policy covers the self-hosted Software. If you separately choose to use the optional Rory Plans cloud service, your data is processed on Rory Plans infrastructure and the Rory Plans, Inc. Privacy Policy applies to that use.
We may update this policy as the Software evolves. Material changes will be reflected here with a revised effective date, and the version history is available in the project’s public repository.
Questions about this policy or RoBrain’s data handling: ade@roryplans.ai.