Xerg is a graph of the systems a team runs, made by Xerg Inc. This page says what Xerg reads, what it keeps and who processes it. Questions and deletion requests go to hello@xerg.ai; security reports to security@xerg.ai.
Signing in
You sign in with a link sent to your email, with Google or with GitHub. Xerg keeps your email address, and the name and profile picture the sign-in method provides; with GitHub, also your username and the GitHub App installations you can see. Sessions are kept in Xerg’s database, and the tokens Google and GitHub issue at sign-in are stored encrypted. Workspace membership (your role, how you joined, when you were last seen) is stored there too.
Workspace access requests and membership decisions are recorded. When you choose to claim workspace ownership through GitHub, Xerg verifies your GitHub identity and organization-owner role using read-only membership access. It keeps the GitHub user and organization identifiers, verification time, and any independently approved workspace role, and rechecks derived ownership at sign-in and after its daily proof expires.
When you sign up, Xerg sends your email address to Cardinal, Xerg’s sales tool, with your first and last name if your account has both, so Xerg can follow up. It isn’t sent if your browser sends a Global Privacy Control signal or you’ve opted out through “Your privacy choices”. To have it removed from Cardinal later, email hello@xerg.ai.
Repositories
With the Xerg GitHub App installed, Xerg reads the file list of the repositories you share and the contents of configuration files: package manifests, deploy and Terraform configs, static Expo app.json files, Cloudflare wrangler configs, CI workflows, Docker Compose files, database schemas and example environment files. From environment files it keeps variable names, never values. Static app.json files can identify Expo’s EAS Update or Heroku; Xerg does not execute them or keep update URLs, project IDs or query strings.
Xerg also reads MCP configuration files inside the repositories you share, to scan configured agents, MCP servers and recognized underlying services. It keeps server names, transport, enabled state, environment variable names, remote endpoint origins (scheme, host and port), recognized package names, and the configuration file evidence. It discards raw command lines and arguments, credential values, HTTP headers, and URL paths, queries and fragments. It does not run configured commands, contact those MCP endpoints, or verify an agent’s access. Personal configuration outside the scanned repository is not read.
It also reads up to 300 source files per repository, none over 100 KB, and scans their import lines, API hostnames and explicit API or SDK calls (including Expo push tokens and browser PostHog initialization), to find the services Xerg recognizes. From those it keeps the service, the file, the line and a signal classification (import, API call, endpoint, OAuth configuration, navigation link or asset load), never the code or call arguments. This is on by default; a workspace owner or admin can turn it off in setup, and Xerg then reads configuration files and environment variable names only. An owner or admin can also leave apps out in setup, and Xerg then reads nothing in their folders.
What it keeps is the graph it derives: names of apps, services, resources and configured MCP agents and servers, how they connect, and which file and line each came from. File contents are not stored. Public repositories scanned from the home page are stored, and scanned again when someone opens one more than 24 hours later; to limit abuse, visitors are counted by a one-way hash of their IP address, never the address itself.
Connected services
When a workspace admin connects a service (for example Cloudflare, Stripe or OpenAI), the key they paste or the token the service issues after sign-in is encrypted with AES-256-GCM and stored for that workspace. Xerg uses it only to read the account: names, regions, status, counts and, for AI providers, spend totals. It never makes changes, never sends a key back to a browser, and deletes it when an admin presses Disconnect.
Keys for coding agents (Xerg’s MCP server) are stored hashed. For each key, Xerg keeps its name, who made it, the name and version of the agent’s MCP client as the client reports it (for example Claude Code 2.1.3), when it was last used, and how many calls it made each month. Never what the agent asked.
What a connection finds is added to the workspace’s graph and is visible to its members. Xerg does not keep customer data held in those services, such as database rows, emails, payments, messages or error events.
Optional production evidence and verification
When a workspace configures production sources, Xerg keeps source definitions, repository commits, deployment identifiers, routing, observation times and coverage gaps. Opt-in collectors can report selected navigation paths and audience labels, documentation and bot-source hashes, runtime service relationships and versions, API contract results, webhook delivery and processing metadata, and declared/observed resource identifiers. Findings record evidence, ownership and review decisions. Collectors discard raw HTML, bot answers, API bodies, cookies and telemetry payloads before submission; source authorization stays on the customer-controlled runner. Screenshots stay on that runner and only their hashes reach Xerg.
If a workspace explicitly enables a canonical documentation source, Xerg also reads and stores the Markdown files listed in that repository’s reviewed context catalog, at an immutable Git revision. Stored document text, titles, owners, hashes and source links are available to authorized workspace members and connected agents. The collector does not follow document links; file and collection limits apply. Do not include secrets or personal records in the catalog. Documentation is guidance and is not independently verified production behavior. Document observations follow the same history retention rules below.
Notion document retrieval and product-definition/claims publication were retired on November 6, 2026. Xerg no longer reads or writes Notion pages. Stored integration and feature-webhook credentials are cleared, sources are disabled, and cached Notion text is unavailable through app and agent endpoints. Existing Notion pages are unchanged. Inactive publication and approval records remain private historical records until their workspace is deleted. Old document observations are inaccessible and follow their existing 90-day retention. GitHub documentation collection and detection of Notion dependencies in scanned code remain supported.
Protected-operation verification records the actor, target, proposed artifact, command digest, policy and evidence revisions, decision, review override and host-reported outcome. Source-only collector credentials and ordinary receipt tokens are stored hashed; an owner-approved receipt is stored encrypted and can only be used during its short validity window. Observation history, refresh events, audit records and verifications are retained for 90 days, except the current observation remains available even when older. Source configuration and findings remain until the workspace is deleted. Production context documentation explains scope, runner setup and enforcement limits.
Optional change reports and local collection
When a workspace enables observed graph updates, Xerg stores source-attributed changes, deployment and configuration comparisons, and reports for 180 days. Underlying observations are kept for 90 days plus the current observation. Reports identify expired supporting detail. Human corrections, ownership notes, scoped retirement reviews, review decisions and notification preferences remain until the workspace is deleted. Clearing the current graph does not delete retained observations, reports or review history.
Optional repository review collection keeps pull-request titles and descriptions, author, reviewer and merger identities, and review decisions. Repository access can also use an explicitly configured Bitbucket Cloud or GitLab.com token, stored encrypted. Agent instruction analysis keeps paths, revisions and extracted references, not full instruction files, and never executes their contents. Gateway analysis keeps explicit alias-to-deployment mappings and model, version, region and deployment-type metadata.
Restricted Kubernetes collection retains allowlisted resource metadata, image identifiers, controller relationships and reconciliation status. It excludes Secrets, environment values, ConfigMap bodies and arbitrary custom-resource specifications. Collection can run locally with bounded file export; uploading an export to hosted Xerg is explicit. The independent local core needs no Xerg account or telemetry, and its graphs and history stay on the local machine unless the operator uploads them.
Retirement reviews record the submitting member, responsible owner, checked scope, usage-evidence links and remaining uncertainties. Report feedback records usefulness, estimated time saved and corrections. Alerts and weekly digests are explicitly subscribed to; they send selected findings and report links through Resend, with membership and subscription checked before delivery. A workspace can pause email delivery independently of collection.
Sharing
A workspace’s graph and connections are visible to its approved members only. Admins and owners invite people by email with a role; an invite works for 7 days and only for that address. GitHub repository access lets people discover a workspace and request access, but does not automatically grant membership or access to its graph. A verified GitHub organization owner can explicitly become a workspace co-owner after acknowledging the seat impact; existing owners keep their access. Admins and owners can approve access requests, change roles, remove people and clear the graph.
Cookies and browser storage
On xerg.ai and in the app:
- CookieYes’s consent cookie keeps the choices you make in the cookie banner. It’s set for xerg.ai and its subdomains, so one choice covers the site and the app. In the EU and UK, Analytics cookies and Advertising are off until you allow them. Outside those regions, Analytics cookies are on, and Advertising is on unless you opt out or your browser sends Global Privacy Control. Global Privacy Control blocks Advertising, not analytics. The round privacy icon at the bottom left of xerg.ai opens the cookie preferences where the banner applies, or a sale/sharing opt-out panel elsewhere. If CookieYes cannot confirm the applicable choices, Xerg uses cookieless analytics and keeps visitor identification off.
- PostHog analytics counts visits: the pages you view and for how long, the links and buttons you click on xerg.ai, the page that sent you, and your browser, device and rough location. When Analytics cookies are not allowed, it sets no cookies and keeps nothing in your browser: PostHog tells visits apart by a one-way hash of your IP address and browser that changes every day. When Analytics cookies are allowed, a PostHog cookie keeps an anonymous ID across xerg.ai and the app, so a visit and a later sign-up count as the same person. Xerg doesn’t record your screen or sessions.
- Cardinal’s website-visitor identification, run by RB2B, only on xerg.ai and only when Advertising is allowed. RB2B also loads LiveIntent’s identity script. These services set cookies and may match your visit to a business profile (such as a name, job title and company) so Xerg can follow up. It never runs when your browser sends a Global Privacy Control signal.
In the app, once you sign in, analytics events carry your Xerg account’s ID and your workspace’s ID, never your email address, name or GitHub username. They record the pages you open (without anything after the “?” in their address) and steps such as installing the GitHub App, creating a workspace, your first graph, connecting an account, adding an evidence source or a production check (its kind only), reviewing a problem (the decision only) and asking Ask Xerg a question. Never the question itself, a source’s repo or hostname, which problem it was or the reason you gave. When Analytics is allowed, they also record which view of the graph you switch to (3D or 2D, and organic, grid or inventory), and nothing about what’s on it. They record that you opened an item’s details, which action you took there (such as refresh, connect, copy or which tab you opened) and which explanation you opened, by kind only: that the item was a service, a repo, an app or a resource, and whether it was connected, detected in code or unrecognized. Never which item, its name or its repo.
With Analytics allowed, the app also records sign-up method and whether you were invited, invitations and joins, later graphs, and successful public-repo explores (whether featured, never the repo name). Checkout and subscription events include the billing interval and seat counts; cancellation and failed-payment events contain IDs, not card details or other billing data. An internal-account flag excludes Xerg’s own activity without sending email addresses.
Xerg stores your latest Analytics choice for activity without a browser. Agent use is counted under a separate agent-key ID, at most once per key per day, only when the key’s creator has allowed Analytics. Billing events sent from Stripe’s webhook require the permission of the person who started that workspace’s checkout. Unknown or denied consent sends neither. Changing Analytics preferences in the app updates this saved choice and stops future events; it does not erase events already received by PostHog. Xerg does not connect its Stripe billing records to PostHog.
Under some US state privacy laws, Cardinal’s visitor identification, and sending Cardinal your details when you sign up, count as sharing personal information for advertising. To opt out of both, use the round privacy icon at the bottom left of xerg.ai: turn off Advertising in the cookie preferences, or choose the sale/sharing opt-out in the panel shown outside the banner regions. You can also turn on Global Privacy Control in your browser. Xerg keeps this opt-out in a cookie shared with the app and in browser storage. If identification already ran, opting out clears the RB2B and LiveIntent identifiers Xerg can access in its own cookies and storage; it cannot delete third-party cookies or information already received by those services.
Xerg also uses:
- Session cookie, to keep you signed in (30 days, renewed while you use Xerg).
- Short-lived cookies that protect sign-in and connection round trips (10 to 60 minutes). Email sign-in uses a 15-minute cookie holding a one-way hash of the requested link, so that browser can finish signing in automatically. Other browsers require confirmation. The cookie is cleared when the link is verified.
- Browser storage for your own choices: in the app, 3D or 2D, the graph’s layout (organic, grid or inventory), a collapsed sidebar, whether the Repos panel is open, which inventory sections are open and the last workspace; on xerg.ai, light or dark, and whether you’ve opted out of Advertising.
- The last 30 Ask Xerg messages, in the tab’s session storage, until the tab is closed.
Who processes data
- Vercel hosts Xerg and runs its servers.
- Neon hosts its Postgres database (workspaces, graphs, encrypted connections).
- GitHub provides sign-in and repository access, and serves profile pictures.
- Google provides sign-in, if you choose it, and serves its profile pictures.
- Resend sends sign-in, invite and explicitly subscribed report emails: it receives the address and the message.
- Stripe handles payment for paid plans: it receives the email of the owner who subscribes, the card details they enter on Stripe’s own page (Xerg never sees them), the workspace’s ID and how many members and agent seats are billed.
- Anthropic runs the model behind Ask Xerg: it receives your question, the conversation’s earlier messages, the workspace’s name and what Ask looks up on the graph.
- PostHog (US) runs Xerg’s analytics on xerg.ai and in the app. It receives the events above, your IP address (for rough location and, without cookies, the daily hash), browser details, and the anonymous ID or, once you sign in, your Xerg account’s ID.
- CookieYes shows the cookie banner and keeps a record of your choice.
- Cardinal, RB2B and LiveIntent support business-visitor identification on xerg.ai, only with Advertising allowed. They receive your IP address, browser details and the pages you visit. Cardinal also receives a new account’s email address and name (see “Signing in”) and may add a business profile to it, such as a job title and company.
- Brandfetch and the Simple Icons CDN serve service logos to your browser. Those requests include the service’s domain name (for example stripe.com) and, like any web request, your IP address.
- Better Stack checks that Xerg is up and runs its status page, status.xerg.ai. The status at the bottom of every page on xerg.ai loads from Better Stack, so, like any web request, it receives your IP address.
- The services you connect receive Xerg’s read requests made with your key or token.
Keeping and deleting
Graphs are kept until an admin or owner clears them or the workspace is deleted. A compact copy of each day’s graph (its items’ names, kinds and statuses, how they link, and saved MCP configuration metadata) is kept for 180 days, even after the graph is cleared. Connections are kept until someone disconnects them. Uninstalling the GitHub App stops repository access. To delete a workspace and everything in it, email hello@xerg.ai from an owner’s address.