Clawvisor

" When you use Clawvisor's LLM proxy, Clawvisor processes prompts, conversation history, tool definitions, tool calls, tool results, model outputs, streaming chunks, and related request/response content only transiently to provide proxy features such as credential substitution, tool-call inspection, task-scope enforcement, human approval, redaction, rewriting, blocking, and audit logging. Clawvisor does not persist raw LLM request bodies, raw LLM response bodies, or full model streams. We may store metadata and redacted audit records, such as provider, model, endpoint, timestamp, status code, streaming flag, tool name, decision, outcome, task ID, agent ID, and sanitized parameters. Redacted audit records are retained as audit logs. Short-lived pending approval state, nonces, and resolver state may be stored temporarily and deleted when the approval, stream, or request no longer needs them. Operational trace/debug logs, when generated, are redacted or sanitized and retained only as needed for service operation, security, support, or compliance, then deleted according to the applicable account deletion or operational log retention process."

" When you self-host Clawvisor, your credentials, audit logs, agent configurations, and all data processed by your instance remain under your control. The cloud relay is enabled by default for local daemon installations, and push notifications are enabled when you pair a mobile device; connection metadata for these services (described in Section 3a) is processed by Clawvisor's infrastructure. We do not collect telemetry or usage analytics from self-hosted instances unless you explicitly opt in. When enabled, anonymous telemetry includes aggregate event counts (tasks created, gateway requests processed, approvals issued) and is not linked to any user identity or credentials."

"Raw credentials are never included in logs. LLM proxy records: Metadata, redacted audit records, short-lived approval state, nonces, resolver state, and operational trace/debug logs related to LLM proxy traffic, as described in Section 2a. Raw LLM request bodies, raw LLM response bodies, and full model streams are not persisted. Notification configs: Telegram bot tokens and chat IDs for approval notifications, if configured. "

" Our marketing site is hosted on Ploy and uses two privacy-focused, cookie-free analytics tools: Ploy's built-in first-party analytics and Plausible Analytics. Neither tool uses cookies or collects personal data, and both are compliant with GDPR, CCPA, and PECR. All data is aggregated and no individual visitors can be identified. Server logs may record IP addresses and request metadata as part of standard web hosting, which are retained for security purposes and automatically purged. The hosted application at app.clawvisor.com uses strictly necessary cookies for authentication and security, such as an HttpOnly refresh-token cookie. These cookies are required to provide the Service and are not used for advertising or analytics."

" Clawvisor is not directed at individuals under the age of 13. We do not knowingly collect personal information from children."

" To the fullest extent permitted by law, Clawvisor, Inc. and its officers, directors, employees, and affiliates shall not be liable for any indirect, incidental, special, consequential, or punitive damages, or any loss of data, revenue, or profits arising from your use of the Service or the Software. This includes damages resulting from actions taken by AI agents, credential exposure, service interruptions, LLM proxy misclassification, modified, delayed, blocked, incomplete, or re-emitted outputs, tool-call handling, provider outages, provider rate limits, account suspensions, provider billing issues, or alleged violations of provider terms or policies. To the extent permitted by applicable law, Clawvisor's total aggregate liability for all claims arising out of or related to the Service or the Software shall not exceed the greater of (a) the amounts you paid to Clawvisor in the twelve (12) months preceding the claim, or (b) one hundred U.S. dollars (US $100)."

" Local daemon installations connect to the Clawvisor cloud relay by default. The following data is transmitted to Clawvisor infrastructure: Connection metadata: Daemon ID, Ed25519 public key, IP address, and connection/disconnection timestamps. Tunnel traffic: All traffic between your daemon and connecting clients is encrypted in transit (TLS). A subset of security-sensitive routes — including agent gateway requests, task management, and connection requests — additionally enforce end-to-end encryption (X25519/HKDF + AES-256-GCM), making their contents unreadable by the relay server. Other routes, including the MCP protocol used by IDE plugins such as Claude Desktop, are protected by TLS in transit but are readable by the relay server. Credentials stored in your local vault are never transmitted through the relay. Authentication: The relay stores your daemon's Ed25519 public key for challenge-response authentication. When you pair a mobile device, push notifications are enabled. The following is transmitted: Device tokens: Required to deliver notifications to your mobile device. Notification payloads: Payloads include event type, task purpose, risk level, and action summary. They do not include credentials, API response bodies, or message contents. Connection metadata and device tokens are retained while the relay connection or push subscription is active and are deleted upon de-registration or account removal."

" Cloud service: API credentials you store are encrypted with AES-256-GCM on our infrastructure. Credentials are decrypted only in memory during request execution and are never logged. Self-hosted: Credentials are stored encrypted on your infrastructure. We have no access to your credentials, data, or server environment. Credentials are decrypted only in memory during request execution and are never logged. Cloud relay: When a daemon connects to the cloud relay, all tunnel traffic is encrypted in transit via TLS. A subset of routes — agent gateway requests, task operations, and connection requests — additionally enforce end-to-end encryption between the client and daemon, making their contents unreadable by the relay. Other routes, including the MCP protocol used by IDE plugins, are protected by TLS only. Credentials stored in the local vault are never transmitted to the relay."

" When you use the LLM proxy, you authorize Clawvisor to process LLM requests, responses, tool definitions, tool calls, tool results, streaming chunks, and related content as needed to provide proxy features. This processing may include inspecting, buffering, redacting, rewriting, holding for approval, blocking, synthesizing approval prompts or audit summaries, and re-emitting LLM requests, responses, tool calls, and tool results. Clawvisor does not persist raw LLM request bodies, raw LLM response bodies, or full model streams when operating the LLM proxy. Raw content is processed in memory or transient buffers only as needed to forward, inspect, redact, rewrite, approve, block, or audit requests and responses. Clawvisor may retain metadata, redacted audit logs, pending approval state, nonces, resolver state, sanitized trace/debug logs, and other operational records needed to provide and secure the service. You remain responsible for the LLM providers, accounts, models, endpoints, credentials, prompts, tools, and outputs you configure or authorize through Clawvisor, including any provider terms, usage policies, rate limits, safety systems, account restrictions, and charges that apply."

" The intent verification feature uses a third-party LLM to check whether agent requests are consistent with approved purposes. This check is best-effort and may not catch all misuse. It is not a substitute for careful task scope design and human oversight. Chain context tracking: When enabled, the output of API calls executed on your behalf — which may include email bodies, message content, file contents, contact details, and other data returned by third-party APIs — is sent partially or in full to the configured LLM provider to extract structural references such as IDs, email addresses, and phone numbers. Credentials are never sent. Self-hosted users select their own LLM provider and are responsible for reviewing that provider's data handling policies. On the cloud service, this processing occurs within Clawvisor's Google Cloud infrastructure. Task risk assessment: When enabled, task purpose descriptions and authorized action scopes are sent to the configured LLM for risk evaluation. This assessment is advisory and does not guarantee that high-risk tasks will be blocked."

" Clawvisor, Inc., a Delaware corporation ("Clawvisor," "we," "us," or "our"), is the data controller for information collected through the cloud service and the clawvisor.com website. Clawvisor is a credential-vaulting gateway for AI agents, available both as a hosted cloud service at app.clawvisor.com ("the Service") and as self-hosted software you run on your own infrastructure ("the Software"). This Privacy Policy covers the clawvisor.com website, the cloud service, and describes what data a self-hosted Clawvisor instance processes."

" When you use the Clawvisor cloud service, we collect, store, or use the following data: Account data: Email address and bcrypt-hashed password for authentication. If you sign in via SAML SSO, your email address is received from your organization's identity provider. Authentication session data: We use strictly necessary authentication cookies, including an HttpOnly refresh-token cookie, to keep you signed in and protect account access. These cookies are not used for analytics, advertising, or cross-site tracking. Credentials you store: API keys and OAuth tokens you add to the vault, encrypted with AES-256-GCM at rest. Agent tokens: Stored as SHA-256 hashes. The raw token is shown once at creation and never stored. Agent tokens may also be issued to third-party applications (such as IDE plugins) through an OAuth 2.1 authorization flow that you explicitly approve. OAuth client registrations: When a third-party application connects via OAuth, we store its client name and redirect URIs. Authorization codes are short-lived and deleted after use. Chain context facts: When intent verification with chain context is enabled, structural references (such as email addresses, IDs, and phone numbers) may be extracted from API responses and stored temporarily to validate follow-up requests within the same task. These facts are automatically deleted when the task completes, expires, or is revoked. Audit logs: Every gateway request is logged with the service, action, sanitized parameters, decision, and outcome. "