Strict Data Sovereignty Zero Model Training on Customer Data

/ ml Privacy Policy & Data Protection Standards

Effective Date: November 8, 2026 · Last Updated: November 8, 2026 · Applies to slashml.ai, app.slashml.ai, api.slashml.ai, SDKs & MCP Plugins

1. Zero Training on Your Data

Your memories, source code, AST indices, debugging traces, robot trajectories, and queries are never used to train, fine-tune, or improve foundation models or shared weights.

2. Zero Third-Party LLM Subprocessors

Unlike wrapper memory layers that forward your prompts to OpenAI or Anthropic on every write, / ml processes memory ingestion natively without sending customer payloads to external LLM APIs.

3. Cryptographic Tenant Isolation

Every workspace, API key, and memory vault (tenant_id, user_id, agent_id, robot_id) is strictly isolated. No cross-tenant recall or shared state ever occurs.

4. Instant Surgical Unlearning

Calling DELETE /v1/memories/{id}, forget_memory, or workspace purge immediately and deterministically removes the data and its index state without lingering background caches.

1. Scope & Privacy-First Architecture

This Privacy Policy governs how SlashML Inc. (“/ ml”, “SlashML”, “we”, “us”, or “our”) collects, processes, protects, and deletes information across our websites (slashml.ai, docs.slashml.ai), cloud console (app.slashml.ai), managed API and remote Model Context Protocol (MCP) endpoints (api.slashml.ai), open-source SDKs (slashml-memory, @slashml/memory, slashml-robotics, @slashml/robotics), and ideological/agent marketplace plugins (Claude, Cursor, OpenAI Codex, Gemini CLI).

We built / ml specifically for engineering organizations, enterprise AI teams, and embodied robotics fleets that require strict intellectual property protection. For Customer Content submitted to our API or MCP servers, SlashML acts strictly as a Data Processor / Service Provider acting on the instructions of the Customer (the Data Controller).

2. Strict Prohibition on Model Training & Secondary Use

We enforce an unconditional technical and contractual prohibition against training on Customer Content:

No Model Training: We never use your stored memories, code repositories, behavioral search indexes, stack traces, prompts, or embodied robotics episodes to train, pre-train, fine-tune, or distill any machine learning model.
No Human Review of Private Vaults: SlashML personnel cannot view or inspect unencrypted Customer Content in your memory vaults unless you explicitly grant temporary support access in writing to debug a specific support ticket.
No Data Brokerage or Advertising: We never sell, rent, share, or monetize Customer Content or personal data with advertisers, data brokers, or third parties.
No Third-Party LLM Forwarding: When you ingest or query memories through api.slashml.ai, your payloads are processed directly by our native memory engine and are never routed to external commercial LLM providers (such as OpenAI, Anthropic, or Google).

3. Information We Collect & Process

Category	What Is Processed	Purpose & Retention
Customer Vault Content	Text memories, scoped rules, code index representations, and robot trajectory episodes explicitly transmitted via API or MCP tools.	Solely to execute memory storage, retrieval, safety gating, and deletion for your authenticated tenant. Retained until deleted by you via API/Console or account termination.
Account & OAuth Identity	Work email address, organization/tenant name, OAuth subject identifier (GitHub, Google, or Enterprise SSO), and billing contact via Stripe.	Authentication, workspace access control, OAuth 2.1 MCP session authorization, and billing administration.
API Credentials	SHA-256 cryptographic hashes and non-sensitive prefixes of API keys (`sml_live_...`).	Request authentication and tenant routing. Raw API keys are shown only once at creation or provisioned directly into your OAuth 2.1 PKCE token exchange.
Operational Telemetry	Request timestamps, HTTP status codes, endpoint latency, and token/operation counts per tenant.	Rate limiting, quota enforcement, abuse prevention, and SLA monitoring. Operational logs never log raw prompt bodies or secret keys.

4. MCP Servers, IDE Extensions & Local Code Indexing

When you connect to / ml via our Remote Streamable HTTP MCP servers (https://api.slashml.ai/mcp or https://api.slashml.ai/robotics/mcp) or local CLI/SDK packages:

OAuth 2.1 with PKCE: Remote MCP authentication uses OAuth 2.1 Authorization Code Flow with S256 PKCE via https://app.slashml.ai/oauth/authorize. You choose the specific tenant vault to authorize, and you can revoke the generated MCP API key at any time in one click from app.slashml.ai.
Scoped Tool Execution: The MCP server only receives parameters explicitly passed when your agent invokes a tool (for example, search_memories, find_bug_candidates, or check_kinematic_safety). It does not passively record your screen, keystrokes, or unrelated local files outside your workspace.
Secret Redaction Gate: Our client SDKs and ingestion pipelines include built-in entropy and credential pattern filters designed to block accidental ingestion of private keys, bearer tokens, or .env secrets before storage.

5. Cryptographic Security & Tenant Isolation

Encryption in Transit: All traffic between clients, MCP hosts, edge workers, and origin memory engines is encrypted over TLS 1.3 with HSTS enforced.
Encryption at Rest: All persistent memory shards, metadata databases, and backups are encrypted at rest using AES-256.
Strict Multi-Tenant Isolation: Every API key is cryptographically bound to a single tenant_id. Memory indices are partitioned per tenant so vector, lexical, and temporal lookups cannot traverse tenant boundaries under any query condition.
Sovereign VPC & Air-Gapped Deployments: Enterprise customers deploying / ml Sovereign inside their own AWS, GCP, Azure VPC, or air-gapped on-premises hardware maintain 100% custody of all storage volumes with zero outbound telemetry to SlashML servers.

6. Deterministic Deletion & Surgical Unlearning

Unlike traditional graph and vector memory services where deleted records remain embedded inside LLM-summarized graph edges or stale HNSW segments, / ml supports immediate, deterministic unlearning:

Single-Record Unlearning: Calling DELETE /v1/memories/{memory_id}, forget_memory, or forget_embodied_episode immediately purges the item from active recall and storage.
Full Tenant / Vault Purge: Calling DELETE /v1/memories/?user_id={scope} or deleting a tenant in app.slashml.ai permanently destroys all memories, indices, and metadata associated with that scope.
Account Deletion Requests: You may delete your account at any time from the Console or by emailing privacy@slashml.ai. All live tenant data is purged immediately upon verification, and encrypted disaster-recovery snapshots age out automatically within 30 days.

7. Minimal Infrastructure Subprocessors & Zero Tracking Cookies

We do not use third-party advertising networks, behavioral tracking pixels, or cross-site analytics cookies on slashml.ai or app.slashml.ai. Cookies are strictly limited to essential, HttpOnly, SameSite authentication session cookies required to keep you signed into the Console.

Our core infrastructure subprocessors are limited to:

Cloudflare, Inc. (United States / Global Edge): Edge TLS termination, DDoS protection, OAuth 2.1 routing, and encrypted D1 account metadata.
Dedicated GPU/Compute Cloud Infrastructure (United States): Encrypted-at-rest execution of the native / ml memory engine.
Stripe, Inc. (United States): PCI-DSS Level 1 payment processing for paid subscriptions. SlashML never stores raw payment card numbers.

8. GDPR, UK GDPR, CCPA/CPRA & International Privacy Rights

Regardless of your geographic location, we extend full data subject rights to all users:

Right of Access & Portability: Export all memories and metadata in machine-readable JSON at any time via GET /v1/memories/ or the Console.
Right to Erasure (Right to be Forgotten): Delete individual memories, scoped namespaces, or your entire tenant instantaneously via API or Console.
Right to Rectification & Restriction: Update or overwrite any memory record or revoke specific API/MCP keys immediately.
Data Processing Addendum (DPA): Enterprise customers requiring a countersigned GDPR / CCPA Data Processing Addendum with Standard Contractual Clauses (SCCs) or a HIPAA Business Associate Agreement (BAA) for Sovereign VPC deployments can request one at privacy@slashml.ai.

9. Contact Our Privacy & Security Team

For privacy inquiries, Data Processing Addendums (DPAs), security audits, or vulnerability disclosures, contact us directly:

Privacy & Data Protection Officer: privacy@slashml.ai
Enterprise Security & Sovereign VPC: enterprise@slashml.ai
Support & General Inquiries: support@slashml.ai