Skip to main content

​
Your AI Should Work for You, Not Against You

IronClaw is built on a simple principle: your AI assistant should work for you, not against you. In a world where AI systems are increasingly opaque about data handling and aligned with corporate interests, IronClaw takes a different approach:

Your Data Stays Yours

All information is stored locally, encrypted, and never leaves your control. No cloud storage, no data harvesting, no vendor lock-in.

Transparency by Design

Open source, auditable, no hidden telemetry or data harvesting. You can inspect every line of code and verify exactly what IronClaw does.

Self-Expanding Capabilities

Build new tools on the fly without waiting for vendor updates. IronClaw can create WASM tools dynamically based on your needs.

Defense in Depth

Multiple security layers protect against prompt injection and data exfiltration. Security is not an afterthought but a core design principle.

​
Trustworthy AI

IronClaw is the AI assistant you can actually trust with your personal and professional life. Every design decision prioritizes:

​
Local-First Data Storage

All data is stored in your own PostgreSQL database with encryption at rest. Your conversations, memory, and credentials never leave your machine unless you explicitly choose to share them.

​
Security as a Foundation

Security isn’t a feature layer - it’s built into the architecture from day one:
  • WASM Sandbox: Untrusted tools run in isolated WebAssembly containers
  • Credential Protection: Secrets are never exposed to tools; injected at the host boundary
  • Prompt Injection Defense: Pattern detection, content sanitization, and policy enforcement
  • Endpoint Allowlisting: HTTP requests only to explicitly approved hosts

​
Open and Auditable

Every component is open source under MIT or Apache 2.0 licenses. You can:
  • Review the source code
  • Build from source
  • Verify no telemetry or tracking
  • Audit security controls
  • Contribute improvements

​
Self-Expanding Without Limits

Unlike closed AI assistants that require vendor approval for new features, IronClaw lets you:
  • Build custom tools dynamically
  • Connect to MCP (Model Context Protocol) servers
  • Add new capabilities without restarting
  • Extend functionality through plugins

​
Design Principles

When security controls are missing or uncertain, IronClaw defaults to denying access rather than permitting it. Empty allowlists deny all requests. Missing authentication fails the request.
Multiple overlapping security layers protect against threats. Even if one layer fails, others provide backup protection. For example, HTTP requests pass through allowlist validation, credential injection controls, and leak detection.
IronClaw treats all components as potentially compromised:
  • WASM tools are sandboxed with capability-based permissions
  • Docker containers run as non-root with dropped capabilities
  • External webhooks require shared secret authentication
  • LLM output is sanitized before execution
Every component receives only the permissions it needs:
  • Per-job bearer tokens for container workers
  • Per-tool credential grants for secrets access
  • Scoped allowlists for network access
  • Read-only filesystems in sandboxes
Full audit trails for security analysis:
  • All tool executions logged
  • LLM calls tracked with costs
  • Job events persisted to database
  • Real-time SSE streaming for monitoring

​
Rust-First Implementation

IronClaw is written in Rust to provide:
  • Memory Safety: No buffer overflows, use-after-free, or data races
  • Performance: Native speed without garbage collection pauses
  • Reliability: Strong type system catches bugs at compile time
  • Single Binary: No runtime dependencies, easy deployment
IronClaw is inspired by OpenClaw but reimplemented in Rust with a security-first design. See FEATURE_PARITY.md for the complete comparison.

​
Architecture Goals

The architecture is designed to achieve:
  1. Always Available: Multi-channel support (REPL, HTTP, Telegram, Slack, web gateway)
  2. Parallel Execution: Handle multiple jobs concurrently with isolated contexts
  3. Persistent Memory: Hybrid search (full-text + vector) for context retention
  4. Self-Repair: Automatic detection and recovery from stuck operations
  5. Background Automation: Cron schedules, event triggers, and webhook handlers
IronClaw prioritizes your control and privacy while maintaining the power and flexibility of a modern AI assistant.