Infrastructure for production AI agents — security, memory, and developer tools.
Rust-first. MCP-native. Offline-first. Zero telemetry.
All tools share a common architecture: single binary, no cloud dependency, composable via the Model Context Protocol.
AgentShield — Security Scanner for AI Agent Extensions
Offline-first, multi-framework security scanner. Detects command injection, SSRF, credential exfiltration, tool poisoning, and supply chain risks in MCP servers, CrewAI agents, and LangChain tools. 12 built-in detectors, 4 adapters, SARIF output, GitHub Action, VS Code extension. Validated against 7 Anthropic reference MCP servers.
brew tap aiconnai/engram && brew install agentshield
agentshield scan ./my-mcp-serverEngram — Persistent Memory for AI Agents
Production-grade memory engine: hybrid search (BM25 + vector), knowledge graphs, 216+ MCP tools, portable .egm snapshots, multi-agent sync, multimodal memory (image/audio/video), gRPC + HTTP transports. Used in production at MBRAS AI Chat.
cargo add engram-core
# or: pip install engram-clientClaude Toolkit — Developer Workflow Tools
Unified CLI for Claude Code — Claude Primer (repo priming) + MAO Orchestrator (multi-agent coordination with intelligent model routing).
ImageKit MCP Server — Media Asset Management
MCP server for ImageKit.io — upload, search, transform, and manage media assets from any MCP-compatible AI agent.
| Language | Package | Install |
|---|---|---|
| Rust | engram-core |
cargo add engram-core |
| Python | engram-client |
pip install engram-client |
| TypeScript | engram-client |
npm install engram-client |
An AI agent is only as reliable as the infrastructure it runs on.
We build the layer below the model: where data is stored, where access is controlled, and where agent behavior can be audited. Every tool we ship:
- Works offline — no cloud account required to get started
- Ships as a single binary — no runtime dependencies
- Speaks MCP — compatible with Claude, Cursor, Windsurf, and any MCP client
- Is open source (MIT or Apache-2.0) or has a free tier
Agent (Claude / GPT / Gemini)
│
├── AgentShield ← scans MCP servers before deployment
│ detects injection, SSRF, exfiltration
│
├── Engram ← stores episodic + semantic memory
│ retrieves context across sessions
│ syncs to S3/R2 for persistence
│
└── MCP Server ← your tools (ImageKit, databases, APIs)
Built in Rust. Tested in production. Shipped as open source.