Privacy Policy
Last updated: October 26, 2026
1. General
1.1 The website agentbeam.com, the Agentbeam local runtime recorder and command-line tools, Studio, the Beam Sentinel and Agentbeam (workspace edition) browser extensions, App monitoring and its SDK, the Beam MCP server, the team library, integrations and alert workflows, the in-product assistant, and our cloud dashboard at app.agentbeam.com (collectively, the “Platform”) are operated by AISOLO Technologies Private Limited [CIN: U62099MH2023PTC403351] (“Agentbeam”, “we”, “us” or “our”), a company incorporated under the Companies Act, having its registered office at 1003, Kamdhenu Commerz, Sector 14, Kharghar 410210, Navi Mumbai, Maharashtra, India.
1.2 Agentbeam is a security and observability layer for AI agents and the applications that use them. It records agent shell commands, file changes, tool calls and network activity, flags risky actions with evidence attached, scans MCP server configurations, SKILL.md files and packages before you run them, checks what people send to AI chat assistants for sensitive data before it leaves the browser, and watches running applications for attacks (“Service”).
1.3 This Privacy Policy applies to anyone who browses the website, installs the local runtime recorder or a browser extension, creates an account, is invited to or connects a device, browser or application to a workspace, or otherwise interacts with us (“you”, “your”, “User”).
1.4 By using, browsing or accessing the Platform, you agree to the collection, use, disclosure and storage of your information as described in this Privacy Policy and our Terms of Service. Where law requires your consent for something, we ask for it separately.
1.5 You may withdraw consent for optional data collection (such as cloud sync or product analytics) at any time by adjusting your settings, disconnecting the device or browser, or writing to us at [email protected]. Withdrawing consent for data required to operate a feature may mean that feature is no longer available to you.
1.6 The Platform is not directed at, and we do not knowingly collect personal information from, individuals under the age of 16.
Our role depends on whose data it is.
For the activity of your people, devices and applications that a workspace sends to us (“Workspace Data”), the organization that owns the workspace decides what is collected and why, and we process it on that organization's behalf (as its processor or service provider). For account, billing, security, website and product-analytics data, we decide the purposes and are the controller. If you are a member of a workspace, the organization's own privacy notice also applies, and requests about Workspace Data are usually best directed to it first.
2. Local-first by default, cloud where you opt in
Agentbeam is architected so that the most sensitive data it handles — the content of agent shell commands, file diffs, and tool-call arguments — is recorded to local files on your own machine by default, and is not transmitted to us unless you connect that machine, browser or application to a cloud workspace. Connecting is the opt-in: it is a deliberate step you (or your organization's administrator) take, and disconnecting stops it.
2.1 Local runtime recorder, CLI and Studio
The runtime recorder and the SKILL.md / MCP config / package scanner write activity logs and scan reports to a local directory on your device (the latest 10,000 events and 500 scan reports are retained). Known credential formats are redacted before they are stored. We do not receive, store, or have access to this data unless you choose to export it or send it to us (for example, when attaching evidence to a support request). A plain scan or extraction preview stays offline. Optional features you enable yourself (see Section 2.8) send selected data to the provider you choose.
2.2 Cloud dashboard: connected devices
Teams that create a workspace on our cloud dashboard can connect their developers' machines to it for policy sync, findings, sessions, skills, usage and fleet-level visibility. Once a machine is connected, the recorder sends each recorded agent action to the workspace: the agent and tool name, the action type, a summary of the action (which can include the shell command, file paths, or the change being made, up to 8,000 characters), timestamps, rule matches, and, where the agent reports them, the model used, the prompts under which actions were taken, and token usage. We use token usage and model to show cost estimates and session tips in the dashboard. If your administrator enables full event payload collection, normalized raw payloads are also stored; turning it off stops future collection but does not delete earlier payloads. Each connected machine is registered with its hostname, operating system, a public key, the Agentbeam CLI version and, where relevant, the person it belongs to. When the scanner runs on a connected machine, the workspace receives the scanned item's name, a hash of its contents, its line count and any findings, but not the file contents. Local recording continues in parallel and remains the source of truth on each machine.
2.3 Skills, the team library and remote requests
A connected device reports the skills it has (name, location type, content hash, risk score and findings). The text of a skill file is not stored by default: an authorized administrator can ask a device to send one SKILL.md file, which is then redacted, cut at 64 KB, shown only to people with device-write access, and audited. Authorized administrators can also ask a device to remove a skill or update the Agentbeam CLI; the device carries out the request at its next poll, only on items in its own inventory, and reports the result. A person can disable these on their device with the documented environment settings. Content you publish to the team library (skills, instruction files, commands, subagents and rules, up to 512 KB per share) is stored in your workspace with its versions, author, scan result and review state, and is visible according to your workspace's sharing and review settings.
2.4 Browser extension (Beam Sentinel, free edition)
The Beam Sentinel browser extension checks prompts and actions on supported AI assistant websites against a local rule set before they are sent. Checks run in your browser; the extension does not transmit the content it inspects to us. It stores a rolling local history of recent checks (enabled/disabled state, counts, and short redacted previews) in your browser's local storage, which never leaves your device.
2.5 Browser extension (Agentbeam, workspace edition)
The Agentbeam browser extension is for organizations with an Agentbeam workspace. It runs only on the supported AI chat websites listed on its Chrome Web Store page. It checks prompts, pasted text, AI responses and file uploads on those sites against the workspace's policy. Text-like files are read and scanned inside your browser and discarded; binary files (such as PDFs, Office documents, images and archives) are not inspected. Checks run in your browser, and depending on that policy the extension may block, warn about, or redact sensitive content before it is sent. No activity is collected until you sign in with your Agentbeam account and approve the browser.
Once connected, the extension sends activity metadata to your organization's workspace at app.agentbeam.com: which supported website was used, the type of action (visit, prompt, paste, upload, response, copy and similar), the time, the outcome (observed, blocked, redacted or allowed once), the types and counts of sensitive data found, the length of the text, the model in use where the site shows it, and the type and size of uploaded files. It does not send page addresses, conversation titles, file names or contents, or the sensitive values it detects. Prompt and response text is not sent unless your administrator turns on redacted text. Even then, the text is redacted in your browser before it is sent and is capped at 4,000 characters.
The extension stores a device credential, the cached workspace policy, and a queue of up to 24 hours of undelivered activity in your browser's local storage. Activity is attributed to your account in the workspace, where your organization's administrators can see it. Administrators can also choose to receive email alerts when a critical finding is blocked. Disconnecting the browser in the extension's settings stops collection, revokes the credential and clears the stored data. If you uninstall the extension while connected, your browser notifies the workspace so it can mark that browser as removed; this notice includes the browser's ID and a one-way hash of its credential, but not the credential itself.
2.6 App monitoring and dependency checks
App monitoring watches an application that an organization chooses to monitor. It is off until a workspace owner or administrator turns it on and creates a token for each app. Lines are matched on the server (or inside the app, if the SDK is used) where monitoring runs; raw log lines are never sent. A finding contains the rule, severity, file, a repeat count and a redacted excerpt of up to 600 characters. Redaction removes credentials, bearer tokens, URL userinfo, email addresses and custom detector matches, but keeps query strings, because in an access log the query is usually the attack.
Findings can describe the people who visit or attack the monitored application: their IP address, country and network, request method, path and status, user agent and the agent it claims to be, whether browser headers were present, whether a request carried a Web Bot Auth signature claim, a Cloudflare bot verdict if the app passes one, and opaque user identifiers the app chooses to report. Only the presence, not the value, of certain headers is checked. Sender addresses can be paused, which may be used to answer them with a refusal. Dependency checks send package names and versions only. The organization that operates the application is responsible for notifying its own visitors and users, and acts as controller for this data; we act as its processor. We use it to provide the monitoring, alerts and analytics. When LLM review is on, redacted finding excerpts (in chunks of up to 10) go to a language model (see Section 5); if you provide your own key, that provider receives them and AgentBeam does not.
2.7 In-product assistant and tips
The assistant in the dashboard sends your question, the page you are on, product documentation and, for roles that can see them, counts and labels about your workspace (not finding details, device hostnames or member emails) to a language-model provider to produce an answer. We record the question (best-effort scrubbed of emails and long numbers, and cut to 300 characters) with your workspace ID and page to improve the documentation. Please do not enter secrets or personal data into it. Session tips and similar recommendations in the dashboard are rule-based and computed from data already in your workspace.
2.8 Optional remote checks you enable
Certain features are off until you configure them, including action judging with Jev (TypeSafe) or with a general model or decision service you choose. When enabled, selected, redacted action data (agent, tool, tool input, command) is sent to the provider or endpoint you configure, under that provider's terms, using a key you supply and which may be stored locally on your device. We do not receive it unless the endpoint is ours. Integrations you connect (Slack, Microsoft Teams, Discord, Google Chat, PagerDuty, email, webhooks) receive the alerts you route to them, which can include finding details and attacker-written text.
2.9 We do not sell the data described in this section, we do not “share” it for cross-context behavioral advertising, and we do not use it to train third-party foundation models.
3. Information we collect
The categories of information we collect depend on how you use the Platform:
| Category | Examples | When it's collected |
|---|---|---|
| Account information | Name, email address, password or SSO identifier, organization name, role, and your answers to the short onboarding questions (such as your role and whether you write or ship code), how you arrived (for example a setup link) | Creating a cloud dashboard account or being invited |
| Workspace and membership | Workspace name and ID, members and roles, invitations and permitted email domains, security contact, owner privacy and collection settings | Using or administering a workspace |
| Billing information | Legal entity name, billing email, tax ID (VAT, GST or EIN), purchase order number, billing address, plan tier, seats and invoices; card details are handled directly by Stripe and not stored by us | Subscribing to a paid plan or editing billing details |
| Connected devices | Hostname, operating system, a public key, Agentbeam CLI version, the person it belongs to, last-seen time, and for browsers a label such as "Chrome on macOS" with a random ID | Connecting a machine or browser to a workspace |
| Synced agent activity and sessions | Agent and tool names, action summaries (which can include commands, file paths and changes), prompts where captured, timestamps, rule matches, session grouping, model and token usage | Using the recorder on a machine connected to a workspace (Section 2.2) |
| Skills, MCP and scan results | Item name, content hash, line count, risk score and findings; a SKILL.md file only on an administrator's request, redacted | Running the scanner on a connected machine (Section 2.3) |
| Team library content | Skills, instruction files, commands, subagents and rules you publish, with author, versions, scan and review state | Publishing to the team library |
| Browser extension activity | Supported AI website, action type, time, outcome, finding types and counts, upload type and size, model; redacted text only if your administrator enables it | Using the Agentbeam browser extension connected to a workspace (Section 2.5) |
| App monitoring data | App name and token hash, findings with rule, severity, outcome, redacted excerpt, visitor IP, country, network, user agent and request details, dependency names and versions, learned outbound hosts, opaque user identifiers | A workspace turns on App monitoring and an app reports (Section 2.6) |
| Integrations and alerts | Encrypted webhook URLs and keys, workflows, recipients, and a delivery log of alert attempts and what the destination answered | Configuring integrations and when alerts fire |
| Audit log | Who changed what and when: invitations, role changes, policy and setting changes, risk reviews, remote requests, token and integration changes; the fields changed but not secret values | Administrative actions in a workspace |
| Assistant questions | Your question (scrubbed and truncated), page and workspace ID | Using the assistant (Section 2.7) |
| Website and dashboard usage | Pages viewed, referrer, approximate location, device and browser type, a random device ID, your user ID once you sign in, product events (such as sign-up, onboarding and connection steps), and masked session recordings | Browsing agentbeam.com or using the dashboard (see Section 6) |
| Communications | Support requests, demo and waitlist forms, newsletter subscription and preferences, attachments you choose to send | Contacting us or subscribing |
| Security and abuse signals | IP address, request metadata, sign-up email checks, rate-limit counters | Protecting the Platform from fraud and abuse |
We do not require the content of your source code, prompts, or agent outputs to operate the local recorder or the Beam Sentinel extension, and the cloud dashboard only receives data from machines, browsers and applications that have been connected to it, as described in Section 2. Source code, prompts and logs you choose to connect can nevertheless contain personal data or confidential material; redaction is pattern-based and not complete.
4. How we use information
- To create and give you access to your account and workspace on the cloud dashboard
- To provide the Service, including fleet visibility, policy sync, sessions and tips, skills review, token usage and spend analytics, App monitoring, alerts and workflows, the team library, and evidence export for organizations that opt in
- To process payments, manage subscriptions and issue invoices
- To send account, billing, invitation and security notifications, and alert emails an administrator configured
- To provide customer support and respond to requests
- To maintain and improve the reliability, security, performance and usability of the Platform, including developing detection rules and threat intelligence from de-identified or aggregate data, and publishing aggregate statistics
- To understand how the website and dashboard are used, including through product analytics and masked session replay, to fix issues and improve onboarding
- To detect, investigate and prevent fraud, spam sign-ups, abuse and security incidents
- To send product updates and marketing communications, only where you have opted in or law otherwise permits, and always with an unsubscribe option
- To enforce our Terms and protect our rights, and to comply with legal obligations and respond to lawful requests from public authorities
5. Disclosure and sub-processors
We do not sell your personal information. We share information only with:
- Your organization. If you use a workspace, its administrators and the roles it assigns can see the data described in Section 2, including activity attributed to you, subject to its owner privacy settings
- Service providers (sub-processors) that help us operate the Platform, under contractual confidentiality and security obligations, listed below
- Destinations you configure, such as chat tools, paging tools, webhooks, email recipients, and AI or decision providers you connect with your own key
- Professional advisors (legal, accounting, insurance) where necessary
- Law enforcement, courts or regulators, where required by law or to protect the rights, safety or property of Agentbeam, our users, or the public
- A successor entity, in the event of a merger, acquisition, financing or asset sale, subject to this Privacy Policy continuing to apply
- Anyone else with your consent or at your direction
Current sub-processors
| Provider | Purpose | Data involved |
|---|---|---|
| Cloud infrastructure and database hosting providers | Host the website, the cloud dashboard, its databases and caches | Account, workspace, device, activity, finding, audit and App monitoring data |
| Stripe | Subscription billing, payment methods, invoices and the billing portal | Billing contact, billing address, tax ID, plan and payment details (card data is held by Stripe, not us) |
| Amazon Web Services (Amazon SES) | Transactional, security-alert and newsletter email delivery | Recipient email address, message content and delivery status |
| Resend | Transactional and account email delivery | Recipient email address, message content and delivery status |
| PostHog | Product analytics and, with inputs masked, session replay on agentbeam.com and the dashboard | Pseudonymous device ID, signed-in user ID, pageviews, product events, masked session recordings |
| Google Analytics | Aggregate website traffic measurement on agentbeam.com | Pageviews, referrer, approximate location, device and browser type |
| OpenAI | Powers the in-dashboard assistant and, unless you supply your own key, optional LLM review of App monitoring findings | Assistant questions and workspace counts and labels; redacted finding excerpts (never raw log lines) |
We will update this list when it changes. Customers can ask for notice of changes at [email protected].
7. Retention and deletion
Locally recorded activity is retained on your own device under your control — you can configure retention or delete it at any time using the Platform's local tooling; we have no access to it and no ability to delete it remotely.
Cloud dashboard account and synced data (including browser extension activity, App monitoring findings, sessions, skills, library content and audit entries) is retained for as long as your organization's account is active unless your organization deletes it sooner, and for up to 45 days after account deletion to allow for recovery from accidental deletion, after which it is permanently deleted, subject to any longer retention required by law (for example, billing records) and to residual backup cycles. Alert delivery logs are kept for 14 days. Where a workspace is self-hosted, retention is the operator's responsibility. Disconnecting a device stops future collection; it does not delete data already synced.
Website and product analytics data, support records and billing records follow the retention periods set out on our Data Rights page.
8. Security
We use technical and organizational measures designed to protect information transmitted to our cloud infrastructure, including encryption in transit, access controls and least-privilege principles, hashing of App monitoring tokens, encrypted storage of integration URLs and keys, role-based access inside workspaces, and an audit log of administrative changes. No method of transmission or storage is 100% secure; if you become aware of a security issue affecting the Platform, please report it to [email protected]. If we learn of a breach affecting your personal information, we will notify you and regulators as required by law.
9. Open-source and self-hosted components
The Agentbeam CLI is open source (AGPL-3.0-only), published at github.com/whyashthakker/beam-cli. If you run a self-hosted or open-source build, this Privacy Policy applies only to the extent you connect that build to our website or cloud dashboard; data you never send us is never in our possession, and for a self-hosted dashboard you are the controller and operator. The license terms for the CLI are published alongside its source.
10. International data transfers
We are headquartered in India and may process data using service providers located in other countries, including the United States. Where we transfer personal data internationally, we rely on appropriate safeguards such as standard contractual clauses, consistent with the requirements described in our GDPR Compliance page.
11. Your rights and choices
Depending on where you live, you may have rights to access, correct, delete, or port your personal information, to object to or restrict certain processing, to withdraw consent, to nominate someone to exercise your rights, and not to be discriminated against for exercising them. See our Data Rights page for the full list of rights and how to exercise them.
California and similar laws. In the past 12 months we have collected the categories in Section 3 and disclosed them for business purposes to the recipients in Section 5. We do not sell personal information or share it for cross-context behavioral advertising, and we do not knowingly sell or share data of people under 16. We use sensitive data only to provide the Service and as permitted by law.
Workspace members. Because your organization controls Workspace Data, we may refer a request about it to your organization.
12. Changes to this policy
We may update this Privacy Policy from time to time, for example when we ship new features. Material changes will be notified via the Platform or by email to registered account holders before they take effect. The “Last updated” date shows the current version. Continued use of the Platform after an update constitutes acceptance of the revised policy.
13. Contact and redressal
Questions, grievances or complaints about this Privacy Policy can be sent to our grievance officer:
- Designation: Grievance Officer
- Email: [email protected]
- Registered office: AISOLO Technologies Private Limited, 1003, Kamdhenu Commerz, Sector 14, Kharghar 410210, Navi Mumbai, Maharashtra, India
We aim to acknowledge grievances promptly and resolve them within one month. For general support, write to [email protected].