Data Handling

Understand what Rake stores locally, what managed AI can receive, and what telemetry excludes.

Rake is a desktop product with local workspace storage and managed AI services. Local project storage does not mean model inference happens locally.

The Data Path

For a managed AI request, the path is:

Rake desktop -> Rake Gateway -> selected model or reranking provider

The Rake Gateway authenticates the desktop, applies account and model policy, routes the request, and records usage needed for credits, billing, reliability, and abuse prevention.

Depending on the selected model and operation, Rake uses services from OpenAI, Anthropic, Google, or Voyage AI. Voyage AI is used for reranking search candidates and is not a chat model in the model picker.

Stored On Your Machine

Rake stores the working copies of these materials in the workspace or Rake application data:

  • project files and project configuration
  • chat history and supporting conversation state
  • datasheet PDFs, extracted text, indexes, and source references
  • PCB source files, extracted board facts, and analysis results
  • retained build, flash, debug, and runtime evidence

You can remove chat sessions, the active datasheet, and PCB context from Rake. Removing local material does not retroactively remove data already processed by a model provider.

Sent For Managed AI Processing

Rake sends the content needed to perform the request. Depending on what you ask and attach, that can include:

  • your prompt and relevant conversation context
  • selected or retrieved source code and project-file excerpts
  • datasheet queries, excerpts, and search candidates
  • PCB or schematic images, extracted text, and board facts
  • tool definitions, tool results, build output, and debug observations
  • generated responses needed to continue a multi-step agent turn

Rake does not send every file in a workspace by default. Agents retrieve and include context as the task requires, and directly attached material is eligible for processing.

Do not attach secrets, credentials, signing keys, or material that your organization does not permit an external AI provider to process.

Gateway And Provider Retention

The Rake Gateway processes request content only while it handles the request. It does not store prompts, source code, tool arguments or results, model replies, or stream transcripts. It keeps the account, device, model, token, credit, billing, and operational records needed to provide the service, including the identity and usage of each request so that a repeated request is not charged twice. A completed request cannot be replayed from the server. Rake keeps the text it already received, and another answer requires a new request.

Model and reranking providers process request content under the terms that apply to Rake Labs' use of their services. Their retention rules can differ by provider and service. Do not assume zero retention unless Rake Labs has given you a written commitment for your plan.

This documentation does not promise zero provider retention or zero provider training. Plan-specific written terms take precedence over this overview.

Telemetry Is Separate

Product telemetry is enabled by default and can be disabled in Rake preferences. Usage, error, and crash categories can also be controlled separately.

Telemetry accepts a bounded set of diagnostic fields such as application version, operating system, workflow name, timing, counts, and error fingerprints. The desktop telemetry boundary rejects project-content fields, including:

  • prompt text
  • source code and file contents
  • raw tool output
  • datasheet contents
  • PCB contents and image data
  • request and response bodies

Telemetry controls do not disable managed AI transmission. If you use managed AI, the request content described above still passes through the Rake Gateway to the selected provider.

Practical Controls

  • Review the context attached to a request before sending it.
  • Keep credentials and private keys outside prompts and project attachments.
  • Remove stale datasheet and PCB context when switching projects or board revisions.
  • Delete local chat sessions that you no longer need.
  • Disable product telemetry or individual telemetry categories in Rake preferences when required by your policy.
  • Contact hello@rakelabs.ai for plan-specific data terms or a data request.