Crate cortexai_agents 

Agent Implementation

High-performance agent execution with ReACT loop (Reasoning + Acting)

Features

• ReACT Loop: Reason -> Act -> Observe cycle for intelligent task execution
• Planning Mode: Optional planning before execution for complex tasks
• Stop Words: Configurable termination triggers
• Tool Execution: Parallel tool execution with timeout handling
• Memory: Conversation history and context management
• Discovery: Dynamic agent discovery via heartbeat/capabilities
• Handoff: Capability-based routing between agents
• Guardrails: Input/output validation with tripwire functionality
• Checkpointing: State persistence for recovery and time-travel
• Multi-Memory: Hierarchical memory (short-term, long-term, entity)
• Agent-as-Tool: Compose agents as callable tools
• Self-Correcting: LLM-as-Judge pattern for automatic quality improvement
• Agent Factory: Runtime agent instantiation and configuration
• Durable Execution: Fault-tolerant execution with recovery
• Structured Sessions: Conversation memory and session management
• Approvals: Human-in-the-loop approval for dangerous operations

Re-exports

pub use agent_tool::*;

pub use approval::*;

pub use checkpoint::*;

pub use discovery::*;

pub use durable::*;

pub use engine::*;

pub use executor::*;

pub use factory::*;

pub use guardrails::*;

pub use handoff::*;

pub use memory::*;

pub use multi_memory::*;

pub use persistence::*;

pub use planning::*;

pub use scope_guard::*;

pub use self_correct::*;

pub use session::*;

pub use sqlite_store::*;

pub use trace::*;

pub use trajectory::*;

pub use vector_store::*;

Modules

agent_tool

Agent-as-Tool

approval

Thin re-export shim to match the approval module name expected by docs/issues. The underlying implementation lives in approvals.rs.

approvals

Human-in-the-loop approval system for dangerous tool executions

checkpoint

Checkpointing

discovery

Agent Discovery Module

durable

Durable Execution

engine

Agent execution engine with ReACT loop

executor

Parallel tool executor with approval support

factory

Agent Factory

guardrails

Guardrails

handoff

Multi-Agent Handoff Mechanism

memory

Agent memory management

multi_memory

Multi-Memory System

persistence

Persistent memory with pluggable storage backends

planning

Planning module for agent task execution

scope_guard

Scope-based security guards for tool execution.

self_correct

Self-Correcting Workflows

session

Structured Sessions

sqlite_store

SQLite-based persistent storage for agents

trace

Execution trace collection for agent runs

trajectory

Agent Trajectory Recording

vector_store

Vector Store and RAG (Retrieval-Augmented Generation) System

Macros

agent_template

Macro for defining templates concisely