Reference
What leaves your computer
Faundr has two halves: the plugin, on your computer, and the dashboard, on this site. This page walks through each part and says what stays on your machine and what gets sent to the dashboard. The dashboard stores data in Faundr's database, and only members of the project's organization can see it.
What stays on your computer
- Environment variable values. Faundr never reads
.envto send values. The Stack and the environment variable check only send names. - Tokens for outside services:
SUPABASE_ACCESS_TOKEN,SENTRY_AUTH_TOKENand your hosting CLIs. They're used directly from your computer and never go to Faundr. - The full code map and your checkpoints. The map lives in
.faundr/mapa/and checkpoints live in your git repository. The dashboard gets the map's structure (see below) and, for checkpoints, only the list: when each was saved and the names of the files that changed afterward. - Conversation text used to measure cost. Insights reads the conversation on your computer and only sends the numbers: tokens, tools used, model.
- The content of your skills. Faundr only reads each skill's header (see below). The skill's instructions never leave your computer.
- Text translated in the dashboard. The Translate button uses Chrome's built-in translator, which runs on your computer.
What goes to the dashboard
| What | What gets sent |
|---|---|
| Session log | Every event in your agent's session: your prompts, the tools the agent used, commands, output, files read or edited and the end of each response. Each piece of text is cut at 4,000 characters, and keys and passwords are replaced with *** before leaving your computer. This includes code excerpts, because reading and editing files are agent actions. Only you can see your conversation. Your team sees a summary. |
| Security, quality and design findings | The title, file, line, rule and a short code excerpt, with keys masked. The checks run on your computer, and your full code never goes along. |
| Command errors | The error message and a short output excerpt, with keys masked, and the command that failed (also masked). |
| Tests | Test names, whether they passed or failed, the failure message and coverage per file. |
| Code map | File and function names, the links between them and the lines where they sit. That's what you see in Code map. The "Hide names" option hides the names on screen. |
| Overview, Stack, Showcase | What the agent wrote when running /faundr:overview, /faundr:stack and /faundr:design-showcase. The Showcase also includes the project's CSS so the replicas look the same as your app. |
| Features, memory, handoff notes | What you and the agent record: titles, tasks, rules, decisions, concerns and handoff notes. |
| Skills | For each skill Faundr reads on your computer (from the repo, your personal ones and enabled plugins): the name, the description (up to 1,000 characters), the source and whether only a person can run it. After /faundr:skills, also the when to use, the topics and the type the agent wrote. The skill's content isn't sent. |
| Connected skill results | When a connected skill runs, the agent sends what it found and fixed: the title, severity, file, line, explanation, impact and how to fix it. Keys in the explanation and the fix are masked. |
| Workflows | The drafts the agent sketches with /faundr:workflow, and the progress of each step while it follows a workflow. |
| Usage and cost | Only numbers: tokens per model, tools and the size of Faundr's context. |
| Live app errors | If you connected them with /faundr:errors-setup: the error message, page, browser, version and the steps before the error. The SDK is configured to send no personal data. The original code for each line is recovered on your computer from the source maps, without sending the code. |
Keys in the session log. Before sending, the plugin replaces the keys and passwords it recognizes with
*** (AWS, GitHub, Anthropic and Stripe keys, passwords in database URLs, lines like password=), and the server checks again. A format no rule knows about yet could slip through. If a key ever shows up in a conversation, rotate it with the provider.What goes to other services
| Where | What | When |
|---|---|---|
| OSV.dev (public package vulnerability database) | The names and versions of your project's packages. | During the security check. |
| npm and PyPI | The name of the package the agent wants to install. | Before an install, to check that the package exists and can be trusted. |
| Supabase | Queries from the database test and the Security Advisor, using your token. | Only when you run /faundr:launch, faundr db-test or faundr security-supabase. |
| Sentry | The import request, using your token. | Only when you run faundr errors-import-sentry. |
| Slack or Discord | Error alerts and the weekly summary. | Only if you set up the webhooks in Errors → Live app. |
| TypeSafe (Jev, behind the fast browser) | The task's goal and, at every step, the page's address and title, its visible text (up to 6,000 characters), its list of buttons and fields, and the values you passed for it to type. Never passwords: Jev only learns that a password was provided. | Only when you or the agent run /faundr:browse (or faundr browse). |
What about the AI?
Everything your agent reads goes to the AI provider you use (Anthropic for Claude Code, OpenAI for Codex), as with any use of that agent. Faundr doesn't change that. What Faundr adds is the session-start context (memory, where you left off, findings) and guidance while the agent works. The AI reviews (/faundr:security, /faundr:quality, /faundr:design) and your skills are run by your own agent, on your own account.
Who can see a project's data
- Members of the project's organization, according to their role. See Organization, team and account.
- Your conversation with the agent is yours alone. Your prompts, the agent's replies and the commands are only visible to you, and the database blocks anyone who tries to read them directly. Your team sees a summary of your sessions: duration, branch, number of prompts, files changed, commits, tasks and the handoff note. See Activity.
- Deleting an organization deletes its projects, along with their memory, sessions and findings.
Where the plugin keeps its files
~/.faundr/config.json: your token (and your Jev key, if you saved one withfaundr browse-key). Only you can read this file.~/.faundr/hook.log: the plugin's error log..faundr.jsonand.faundr/: inside your project. You can keep.faundr/out of git..faundr.jsoncan be committed.