Skip to main content

Data, privacy and security

Before working on a closed-source project, decide which files the agent is allowed to read and change, which external tools it can use, and which data may be sent to the selected model.

What data Explyt processes​

The User Agreement treats the following as user data: source code, configurations and settings, chat history, logs and stack traces, test results, build and static analysis reports, dependency information, diffs and pull requests, task descriptions, requirements, comments, custom rules, templates and connection parameters. This category also covers any other content you pass through the Platform or give it access to in your development environment.

Explyt does not analyze the whole codebase with a blanket scan. What ends up in the context depends on the feature you use and the rights you granted. Processing may involve the selected and related files, the project structure, build metadata, dependencies, code changes, logs, test and analysis results, tasks, requirements, comments, and data from connected external sources.

When you connect VCS, issue trackers, CI/CD, MCP and other external services, you control their rights and credentials, and you are responsible for the legality of transferring the data.

Request route​

The general route stated in the User Agreement:

Plugin → AI Server → LLM

The plugin sends user data to the AI Server, which builds the prompt and passes it together with the required data to the LLM to obtain a result.

For Personal, Community and Enterprise, check the exact route, the LLMs used and the processors involved with support or your administrator before sending sensitive data. Ask them about retention terms, deletion timelines, whether materials are used to improve models, and which telemetry settings are available.

Technical access restrictions​

Explyt provides several independent mechanisms:

  • Edit Scope limits the files the agent can change in the current task;
  • .writeignore permanently forbids writing and creating files under the listed paths, but not reading them;
  • .readignore forbids reading the listed paths, but does not replace a write ban;
  • to close a path completely, add it to both files;
  • the Tools menu controls tool availability and confirmation prompts;
  • permissions restrict access to actions and external paths.

Text in AGENTS.md, Rules or a message to the agent guides its behavior but does not create a technical ban. Rules in the ignore files use gitignore-like patterns, with no guarantee that the entire .gitignore syntax is supported. For secrets, list the path in both ignore files, and grant external services only the rights they need.

Before working with sensitive data, separately check the limitations of attachments, MCP servers, IDE logs and third-party tools. Do not rely on the ignore files as a universal restriction across all channels.

Technical and analytics data​

Alongside the data you deliberately send to the agent, the plugin collects usage statistics and telemetry in order to improve the plugin and the Services. This is separate from your source code and chat content, and it applies even on plans where the Company does not process user data.

The authoritative list of what is collected, the purposes it is collected for, and the retention periods is in the Privacy Policy; the User Agreement describes how it relates to your plan. If the exact scope matters to your organization, confirm it with support or your administrator before sending sensitive data.

Local Memory Bank data​

Memory Bank stores Markdown files in the project's .explyt/memory/ folder. When memory is enabled, a compact index is available to the agent, and relevant entries can be opened and added to the request context. Do not keep secrets, credentials or personal data there. The master switch is in Settings → Tools → Explyt → Memory Bank.

When the agent opens a Memory Bank entry for a request, its contents may travel the general processing route. Storing the file locally does not mean its contents are processed locally.

MCP and secrets​

An MCP server gets the rights of its own account and can reach external data. Keep tokens in environment variables and use the ${VARIABLE_NAME} placeholder in the configuration. Substitution is supported only in the args items of a STDIO server and in the headers values of HTTP/SSE servers. After connecting, start with a read-only operation. For details, see MCP and MCP troubleshooting.

If this reference page disagrees with the legal documents, the current version of the legal document prevails.