Crate sombrax_agentic_core 

SombraX Agentic Core (sombrax_agentic_core)

SombraX Agentic Core (SAC) is a Rust library for LLM agent orchestration with content-modifying hooks, MCP tool integration, cross-agent communication, and context optimization.

Features

• Content-Modifying Hooks: Intercept and modify messages before/after LLM calls
• MCP Tool Integration: Connect to MCP servers for tool discovery and execution
• Cross-Agent Communication: Registry for agent discovery and invocation
• Context Optimization: Automatic context management for long conversations
• OpenTelemetry: Built-in observability with tracing and metrics
• LLM Providers: OpenAI, Anthropic, ZAI, Cerebras, OpenRouter (sombrax_agentic_core::providers)
• Agent Tools: File, shell, web, and task tools (sombrax_agentic_core::tools)

Quick Start

use sombrax_agentic_core::prelude::*;

#[derive(Clone)]
struct LoggingHook;

impl Hook for LoggingHook {
    async fn pre_completion(
        &self,
        message: Message,
        _history: &[Message],
        ctx: &mut HookContext,
    ) -> HookResult<Message> {
        println!("[{}] Processing message", ctx.request_id);
        Ok(message)
    }
}

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = OpenAIClientBuilder::new(&std::env::var("OPENAI_API_KEY")?).build();
    let model = client.completion_model_adapter("gpt-4");

    let agent = AgentBuilder::new(model)
        .preamble("You are a helpful assistant.")
        .hook(LoggingHook)
        .build();

    let response = agent.prompt("Hello!").await?;
    println!("{}", response.content());
    Ok(())
}

Re-exports

pub use agent::Agent;

pub use agent::AgentBuilder;

pub use agent::AgentWrapper;

pub use agent::ExecutionStats;

pub use agent::PromptResponse;

pub use context::HookContext;

pub use context::OptimizationConfig;

pub use context::SharedContext;

pub use error::CompletionError;

pub use error::HookError;

pub use error::HookStage;

pub use error::ToolError;

pub use extractor::build_extractor;

pub use extractor::ExtractorBuildError;

pub use extractor::ExtractorError;

pub use extractor::ExtractorWrapper;

pub use hook::builtin::ValidationHook;

pub use hook::builtin::WorkspaceBoundaryHook;

pub use hook::Hook;

pub use hook::HookChain;

pub use hook::ToolCallDecision;

pub use message::validate_tool_result_ids;

pub use message::Message;

pub use message::ValidationError;

pub use provider::CompletionModel;

pub use provider::CompletionRequest;

pub use provider::CompletionResponse;

pub use provider::Usage;

pub use providers::build_agent;

pub use providers::build_agent_with_options;

pub use providers::AgentBuildError;

pub use providers::AgentBuildOptions;

pub use providers::LlmConfigLike;

pub use providers::ProviderType;

pub use providers::ProviderTypeError;

pub use retry::ResponseValidation;

pub use retry::RetryConfig;

pub use retry::ValidationResult;

pub use tool::McpToolSource;

pub use tool::StdioMcpClient;

pub use tool::Tool;

pub use tool::ToolDefinition;

pub use tool::ToolDyn;

Modules

agent

Agent module

context

Context types for hooks and agents

error

Error types for the Agent Hook Library

experience

Experience-based learning system for agent self-improvement. Experience-based learning system for agent self-improvement.

extractor

Extractor module for structured data extraction from LLM responses.

hook

Hook trait and chain execution

message

Message types for conversation content

prelude

Prelude module for convenient imports

prompt

System prompts as first-class on-disk assets with a name-based resolution ladder — analog of skill for persona prompts. System prompt assets — first-class, on-disk prompt files with a deterministic name-based resolution ladder.

provider

LLM Provider abstractions

providers

LLM Providers

retry

Retry configuration for agent execution with exponential backoff.

runs

Pluggable pipeline / bundle / job runtime.

skill

Skill system for discovering and loading user-defined agent skills.

telemetry

Telemetry module for observability

tool

Tool trait and definitions

tools

Agent Tools

Macros

completion_span

Create a tracing span for a completion request

hook_execution_span

Create a tracing span for a hook execution

hook_span

Create a tracing span for a hook chain execution

tool_call_span

Create a tracing span for a tool call