Crate ironclaw 

NEAR AI Agentic Worker Framework

An LLM-powered autonomous agent that operates on the NEAR AI marketplace.

Architecture

┌─────────────────────────────────────────────────────────────────────────────────┐
│                              User Interaction Layer                              │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐                         │
│  │   CLI    │  │  Slack   │  │ Telegram │  │   HTTP   │                         │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └────┬─────┘                         │
│       └─────────────┴────────────┬┴─────────────┘                               │
└──────────────────────────────────┼──────────────────────────────────────────────┘
                                   ▼
┌──────────────────────────────────────────────────────────────────────────────────┐
│                              Main Agent Loop                                      │
│  ┌────────────────┐  ┌────────────────┐  ┌────────────────┐                      │
│  │ Message Router │──│  LLM Reasoning │──│ Action Executor│                      │
│  └────────────────┘  └───────┬────────┘  └───────┬────────┘                      │
│         ▲                    │                   │                               │
│         │         ┌──────────┴───────────────────┴──────────┐                    │
│         │         ▼                                         ▼                    │
│  ┌──────┴─────────────┐                         ┌───────────────────────┐        │
│  │   Safety Layer     │                         │    Self-Repair        │        │
│  │ - Input sanitizer  │                         │ - Stuck job detection │        │
│  │ - Injection defense│                         │ - Tool fixer          │        │
│  └────────────────────┘                         └───────────────────────┘        │
└──────────────────────────────────────────────────────────────────────────────────┘

Features

• Multi-channel interaction - CLI, Slack, Telegram, HTTP webhooks
• Parallel job execution - Run multiple jobs with isolated contexts
• Pluggable tools - MCP, 3rd party services, dynamic tools
• Self-repair - Detect and fix stuck jobs and broken tools
• Prompt injection defense - Sanitize all external data
• Continuous learning - Improve estimates from historical data

Re-exports

pub use config::Config;

pub use error::Error;

pub use error::Result;

Modules

agent

Core agent logic.

app

Application builder for initializing core IronClaw components.

boot_screen

Boot screen displayed after all initialization completes.

bootstrap

Bootstrap helpers for IronClaw.

channels

Multi-channel input system.

cli

CLI command handling.

config

Configuration for IronClaw.

context

Per-job context isolation and state management.

db

Database abstraction layer.

document_extraction

Document text extraction pipeline.

error

Error types for IronClaw.

estimation

Cost, time, and value estimation with continuous learning.

evaluation

Success evaluation for completed jobs.

extensions

Lifecycle management for extensions: discovery, installation, authentication, and activation of channels, tools, and MCP servers.

history

History and persistence layer.

hooks

Lifecycle hooks for intercepting and transforming agent operations.

llm

LLM integration for the agent.

observability

Observability subsystem: trait-based event and metric recording.

orchestrator

Orchestrator for managing sandboxed worker containers.

pairing

DM pairing for channels.

prelude

Re-export commonly used types.

profile

Psychographic profile types for user onboarding.

registry

Extension registry: metadata catalog for tools and channels.

safety

Safety layer for prompt injection defense.

sandbox

Docker execution sandbox for secure command execution.

secrets

Secrets management for secure credential storage and injection.

service

OS service management for running IronClaw as a daemon.

settings

User settings persistence.

setup

Interactive setup wizard for IronClaw.

skills

OpenClaw SKILL.md-based skills system for IronClaw.

tenant

Compile-time tenant isolation.

timezone

Timezone resolution and utilities.

tools

Extensible tool system.

tracing_fmt

Truncating terminal writer for tracing.

tunnel

Tunnel abstraction for exposing the agent to the internet.

util

Shared utility functions used across the codebase.

webhooks

Generic webhook ingress for tools.

worker

Worker mode for running inside Docker containers.

workspace

Workspace and memory system (OpenClaw-inspired).