1. Overview & Core Commitments
This Privacy Policy outlines how Meld collects, stores, processes, and protects your information when you access or use the Meld platform, applications, APIs, Model Context Protocol (MCP) bridges, and dedicated agent workstations.
- Customer Data Ownership: You retain 100% intellectual property and ownership over all customer prompts, connected tools data, spreadsheets, and agent outputs.
- Zero Training Guarantee: Your data is never used to train, retrain, fine-tune, or improve public foundational LLMs.
- Deterministic Runtime Isolation: Every agent computer operates in an isolated, sandboxed environment with strict egress filtering and approval checkpoints.
2. Information We Collect
We collect information in three ways: information you provide directly, information retrieved through authorized third-party integrations, and automated technical telemetry.
A. Information You Provide
Account registration details (name, corporate email, password hash, team size, company name, billing coordinates), room messages, instructions provided to agents, and feedback given during human approval checkpoints.
B. Connected Tool & Stack Data (OAuth & MCP)
When you connect external services—such as Slack, Microsoft Teams, Google Workspace, GitHub, Linear, HubSpot, or Okta—we receive encrypted OAuth tokens and fetch only the resources scoped by your administrator (e.g. specific channels, pull requests, spreadsheet sheets, or CRM records).
C. Sandboxed Workstation Telemetry
Diagnostic performance logs, browser navigation URLs within the sandboxed agent browser, command execution exit codes, and resource utilization metrics necessary to maintain workstation stability and security.
3. How We Use Your Data
We strictly utilize your data to operate, orchestrate, and secure the Meld platform, including:
- Executing agent tasks assigned by team members across your connected tools.
- Indexing entity relationships, vendor rules, and team preferences into your private compounding memory graph.
- Enforcing human-in-the-loop approval gates before agents perform state-mutating actions (e.g. sending cold emails, modifying financial ledgers, or merging pull requests).
- Preventing malicious misuse, verifying identity, and monitoring anomalous API usage.
4. Zero-Training AI Guarantee
Meld does not use your organization's inputs, code, chat streams, memory graphs, or outputs to train or improve any public or shared machine learning models. Our commercial agreements with underlying model providers (e.g. Anthropic, OpenAI, Google Cloud Vertex AI) explicitly include zero-day data retention and zero-training provisions for enterprise API traffic.
5. Sandboxed Agent Computers & Isolation
Unlike traditional chatbots that execute prompts directly on shared serverless workers, Meld agents operate inside isolated containerized computer environments (macOS/Linux userland micro-VMs).
- Micro-VM Perimeter: Each agent session runs in a distinct virtual machine boundary preventing cross-tenant memory access.
- Credential Vaulting: Authentication credentials and API keys are stored in hardware security modules (HSM) and injected only into runtime processes via short-lived ephemeral proxies.
- Restricted Egress: Network egress from agent sandboxes is restricted to whitelisted domains and customer-authorized endpoints.
6. Connected Stack Integrations & OAuth
Meld uses standard OAuth 2.0 and PKCE protocols to interface with your software stack. When you authorize a connector:
- We only request least-privilege permissions required to fulfill the agent's assigned duties.
- Tokens are encrypted at rest using AES-256-GCM.
- You may revoke any integration at any time directly through your Meld Workspace Settings or via the provider's native security console (e.g. Google Security Permissions, Slack App Management).
8. Data Retention, Export & Erasure
We retain customer data for the duration of your active workspace subscription. Upon workspace termination or written request:
- You may export all conversation logs, memory graphs, and task artifacts in standard JSON/CSV formats.
- All customer data, encryption keys, and agent container snapshots are permanently purged from active production stores within thirty (30) days and from offline backup sets within sixty (60) days.
9. Security, Encryption & SOC 2
Meld maintains an enterprise Information Security Program audited under American Institute of Certified Public Accountants (AICPA) SOC 2 Type II standards.
- Encryption in Transit: TLS 1.3 with forward secrecy enforced across all public endpoints and inter-service channels.
- Encryption at Rest: All customer databases, memory vector graphs, and volume attachments are encrypted with AES-256.
- Vulnerability Management: Continuous static and dynamic analysis, automated dependency scanning, and annual third-party penetration tests.
10. Global Privacy Rights (GDPR & CCPA)
Depending on your geographic location, you possess specific statutory rights regarding personal data under the EU General Data Protection Regulation (GDPR) and California Consumer Privacy Act (CCPA/CPRA):
- Right of Access & Portability: Request a copy of all personal data held about you in a machine-readable format.
- Right to Rectification & Erasure: Correct inaccuracies or request immediate deletion of personal data (“Right to be Forgotten”).
- Right to Restrict or Object: Restrict or object to specific processing categories.
- Zero Sale of Personal Data: Meld has never sold personal data and will never sell personal information to data brokers or advertising exchanges.
11. Contact & Data Protection Officer
If you have questions, feedback, or requests regarding this Privacy Policy or our data protection measures, please reach out to our security and legal team: