A single optimized Rust binary. Eleven crates, zero IPC overhead, one SQLite database. Heuristic model routing, 3-level semantic cache, zero-trust agent-to-agent communication, and built-in DeFi yield.

11Crates|28Tables|597Tests|~32KLOC|~15MBBinary
$Quick Install
bash <(curl -fsSL https://roboticus.ai/install.sh)

Core Capabilities

Heuristic Model Routing

Complexity-based heuristic routes queries to optimal models. Rule-based fallback chain with circuit breakers per provider.

Zero-Trust A2A

Mutual ECDSA auth, ECDH session keys, AES-256-GCM encryption. Forward secrecy, per-message HMAC, rate limiting.

DeFi Yield

Idle USDC earns 4-8% APY via Aave/Compound on Base. Auto-withdraw when balance drops below threshold.

4-Layer Injection Defense

Regex gatekeeping, HMAC trust boundaries, output validation, and behavioral anomaly detection.

3-Level Semantic Cache

Exact hash, embedding cosine similarity, and deterministic tool TTL. 15-30% cache hit rates.

Dual-Format Skills

Structured TOML tool chains + instruction Markdown skills. Hot-reload, safety scanning, SHA-256 change detection.

vs OpenClaw

Ironclad replaces a 3-process, multi-language stack with a single Rust binary

DimensionOpenClawIronclad
Architecture3 separate processes (Node.js, Python, TypeScript)Single Rust binary
LanguagesNode.js + Python + TypeScript + GoRust (one language, one toolchain)
Memory usage~500 MB (3 processes)~50 MB (1 process)
Proxy latency~50ms (Python aiohttp)~2ms (in-process, persistent pool)
Cold start~3s (Node.js) + ~2s (Python)~50ms
Binary size~200 MB (node_modules + pip)~15 MB static binary
Supply chain500+ npm + pip packages~50 auditable crates
Database5 storage layers (JSONL, PostgreSQL, SQLite, JSON, MD)1 unified SQLite (28 tables, WAL)

Cost Optimization

Heuristic model routing
Route simple queries to cheap/local models
60-85% on routed queries
Semantic caching
Avoid re-inferring semantically identical prompts
15-30% cache hit rate
Progressive context
Load only what complexity requires (2K vs 16K)
40-60% input token reduction
Connection pooling
Eliminate TCP setup overhead
~50ms per request
Yield engine
Earn interest on idle USDC
4-8% APY

Quick Start

# Build from source
$ cargo build --release

# Run with config
$ ./target/release/ironclad-server --config ironclad.toml

# Or use the CLI
$ ironclad init
$ ironclad setup
$ ironclad serve
Minimal ironclad.toml:
[agent]
name = "MyAgent"
id = "my-agent"

[server]
port = 18789

[models]
primary = "ollama/qwen3:8b"