Enables applications to scan model replies for leaks, sanitize output, authorize model-proposed actions with signed single-use grants, and optionally score prompt injection in a way that can only tighten enforcement decisions.
Exposes an OpenAI- and Anthropic-compatible HTTP, SSE, and stdio gateway that wraps multiple subscription CLIs, adding prompt-injection defense, PII redaction, cost-aware routing, and reasoning-trace capture for MCP-compatible clients like Claude Desktop and Cursor.
OpenAI-compatible proxy with PII scrubbing, audit logging, and model routing. Drop-in replacement — set OPENAI_BASE_URL to point to this server. Supports GPT-4o, Claude, Gemini, and
Llama via single endpoint.
Enables LLMs to propose UPDATE/DELETE SQL that is run in a transaction, measured, and rolled back, requiring human approval before applying to prevent unauthorized changes.
Provides a governance gateway for LLM and agent traffic via MCP, blocking prompt injection, redacting PII, enforcing per-tenant token budgets, and caching repeat questions semantically. Enables AI agents to route model calls through the same guardrails instead of calling providers directly.
Exposes LLM output validation (schema compliance and prompt-injection detection) as an MCP tool, enabling any MCP-compatible agent to apply real-time guardrails without importing the codebase.
Provides a validated MCP server with strict schema-checked tools, alongside security gateway filtering, streaming PII redaction, and rate-limited model fallback routing.
Enables MCP-compliant clients to sanitize prompt payloads by deterministically masking and reversibly redacting PII before they reach LLMs, returning structured output and telemetry.
Enables agents to detect and redact PII and secrets in text, HTML, and PDF documents before they reach an LLM, with reversible placeholder, mask, or hash redaction, and to list supported detectors and redaction styles.
Enables cloud-based MCP clients to delegate privacy-sensitive or large-output tasks to a local LLM, returning sanitized placeholders or digested summaries so private data never leaves the user's network.
AI-powered dependency vulnerability and breaking change analyzer that scans dependencies, identifies vulnerabilities via OSV.dev, and uses AI to assess real impact and suggest fixes.
Enables a cloud orchestrator to query a local Ollama model over Tailscale or a LAN as a side-channel for drafts, summaries, and other helper output, with tools to list available models and to redact text. Answers leaving the server are passed through output-only DLP that masks card numbers, IBANs, emails, phone numbers, API keys, and similar secrets before they reach the cloud.
Enables AI agents to use vault-stored credentials without ever exposing plaintext secrets to model context, transcripts, or telemetry. It decrypts AES-256-GCM envelopes locally via WebCrypto, resolves opaque secret handles into subprocess environments and outbound HTTP requests, issues auto-zeroizing time-bounded leases, and scrubs echoed credentials from output.
Enables LLMs to read a user-owned, local-first memory store kept in a portable folder (USB, SSD or laptop), with per-profile permissions limiting what each client can see and access cut off when the pocket is locked or unplugged. Also lets models propose new memories for the user to review and approve, keeping provenance and keeping canonical memory free of direct model writes.
Enables software authors and security teams to harden their own release artifacts — JS/TS bundles, Electron ASAR, WASM, Python, native binaries, SQLite, GGUF/Safetensors and Android DEX — against AI reverse-engineering agents through an audit → protect → verify toolchain exposed as 31 MCP defense tools. Coding agents can orchestrate format-native shielding, source-level logic weaving, string encryption, scope binding and structural counter-REA verification directly from a prompt, while keeping protected builds runtime-equivalent.
A security-hardened, read-only MCP server that authenticates service and user identities, enforces factory-level access grants, and audit-logs every tool call before dispatching read-only plant overview queries.
Enables querying the MITRE ATT\&CK framework for adversarial tactics, techniques, mitigations, and detection methods through natural language, supporting both ID-based and fuzzy name-based searches.