Crate cognisagent

Expand description

§cognisagent

Batteries-included, high-level agent framework built on cognis and cognisgraph. Provides zero-boilerplate agent creation with pluggable middleware and backends.

§Overview

The primary entry point is create_deep_agent, which constructs a compiled LangGraph CompiledStateGraph with middleware hooks around model and tool invocations.

§Quick Start

use cognisagent::config::DeepAgentConfig;
use cognisagent::create_deep_agent;

let config = DeepAgentConfig::default();
let graph = create_deep_agent(config).unwrap();
let result = graph.invoke(serde_json::json!({"messages": []})).await.unwrap();

§Middleware

The middleware::Middleware trait provides before/after hooks for model calls and tool executions. Built-in middleware:

middleware::filesystem::FilesystemMiddleware – file read, write, list, glob, grep
middleware::memory::MemoryMiddleware – inject persistent memory into context
middleware::subagent::SubAgentMiddleware – isolated sub-agent delegation
middleware::summarization::SummarizationMiddleware – context window management
middleware::skills::SkillsMiddleware – custom skill loading
middleware::patch_tool_calls::PatchToolCallsMiddleware – tool call correction
middleware::rate_limiter::RateLimiterMiddleware – token bucket rate limiting
middleware::logging::LoggingMiddleware – structured logging with redaction
middleware::context::ContextMiddleware – dynamic context injection
middleware::planning::PlanningMiddleware – plan-then-execute with step tracking

§Tool Registry

The tool_registry::ToolRegistry provides centralized tool management with permission levels, call counting, enable/disable control, and filtering.

§Backends

The backends::Backend trait abstracts session state persistence:

backends::StateBackend – in-memory (default)
backends::FilesystemBackend – local disk storage
backends::SandboxBackend – isolated execution with resource limits

§Events

The events module provides a typed event bus system with handlers and lifecycle tracking for observing and reacting to agent execution events.

§Conversation

The conversation module provides a conversation manager with context windowing, message history management, and export/serialization support.

§Presets

The presets module provides ready-made agent configurations via a preset registry and customizer pattern for common agent types.

Re-exports§

pub use agent::create_deep_agent;

Modules§

agent: Deep Agent factory — builds a LangGraph CompiledStateGraph with middleware hooks.
backends: Backend system for persisting agent state.
builder: Comprehensive agent builder with fluent API, templates, and directory.
collaboration: Collaboration protocols for deep agents.
communication: Inter-agent communication system for message passing, shared state, and coordination.
config: Configuration for the Deep Agent.
context: Context management for deep agents.
conversation: Conversation manager for orchestrating multi-turn interactions.
evaluation: Evaluation and benchmarking framework for DeepAgents.
events: Event bus system for agent lifecycle observability.
factory: Agent factory for creating, managing, and tracking agent instances.
health: Health check and diagnostics system for agent components.
lifecycle: Lifecycle management for deep agents.
logging: Structured logging system for DeepAgents execution.
memory: Memory system for DeepAgents with short-term, long-term, and indexed memory.
middleware: Middleware system for Deep Agents.
multi_agent: Multi-agent orchestration patterns for coordinating teams of agents.
observability: Observability and diagnostics for agent execution.
orchestrator: Tool orchestrator for coordinating multiple tool executions with dependency resolution, parallel execution, and result aggregation.
planning: Planning capabilities for deep agents.
plugins: Plugin system for extending agent capabilities at runtime.
presets: Pre-configured agent presets for common use cases.
rate_limiting: Advanced rate limiting for agent operations.
recovery: Error recovery system for agents.
sandbox: Sandbox backend for isolated agent execution.
session: Session management system for tracking agent execution sessions.
skills: Skill system for loading and managing reusable agent skills.
summarization: Summarization middleware for context window management.
telemetry: Telemetry and observability system for tracking agent execution metrics, token usage, and performance.
tool_registry: Tool registry for managing tool availability, permissions, and discovery.
tooling: Built-in tooling system for DeepAgents.
workflow: Workflow engine for multi-step agent workflows with dependency resolution.