Crate agents_sdk

Expand description

§Rust Deep Agents SDK

High-performance Rust framework for composing reusable “deep” AI agents with custom tools, sub-agents, and prompts.

§Quick Start

[dependencies]
agents-sdk = "0.0.1"  # Includes toolkit by default

use agents_sdk::{ConfigurableAgentBuilder, OpenAiConfig, OpenAiChatModel};
use agents_core::persistence::InMemoryCheckpointer;
use std::sync::Arc;

    let config = OpenAiConfig::new(
        std::env::var("OPENAI_API_KEY")?,
        "gpt-4o-mini"
    );

    // Create the model
    let model = Arc::new(OpenAiChatModel::new(config)?);

    // Build an agent
    let agent = ConfigurableAgentBuilder::new("You are a helpful assistant.")
        .with_model(model)
        .with_checkpointer(Arc::new(InMemoryCheckpointer::new()))
        .build()?;

    // Use the agent
    use agents_sdk::state::AgentStateSnapshot;
    use std::sync::Arc;

    let response = agent.handle_message(
        "Please greet Alice using the greet tool",
        Arc::new(AgentStateSnapshot::default())
    ).await?;
    println!("{:?}", response);

§Features

toolkit (default): Includes agents-toolkit with built-in tools
aws: Includes AWS integrations
redis: Redis-backed state persistence
postgres: PostgreSQL-backed state persistence
dynamodb: DynamoDB-backed state persistence (AWS)
persistence: Grouped feature for Redis + PostgreSQL
aws-full: Grouped feature for AWS + DynamoDB
full: Includes all features

§Installation Options

# Default installation with toolkit
agents-sdk = "0.0.1"

# Core only (minimal installation)
agents-sdk = { version = "0.0.1", default-features = false }

# With specific persistence backend
agents-sdk = { version = "0.0.1", features = ["redis"] }
agents-sdk = { version = "0.0.1", features = ["postgres"] }
agents-sdk = { version = "0.0.1", features = ["dynamodb"] }

# With AWS integrations
agents-sdk = { version = "0.0.1", features = ["aws-full"] }

# Everything included
agents-sdk = { version = "0.0.1", features = ["full"] }

§Persistence Examples

§Redis Checkpointer

use agents_sdk::{RedisCheckpointer, ConfigurableAgentBuilder};
use std::sync::Arc;

let checkpointer = Arc::new(
    RedisCheckpointer::new("redis://127.0.0.1:6379").await?
);

let agent = ConfigurableAgentBuilder::new("You are a helpful assistant")
    .with_checkpointer(checkpointer)
    .build()?;

§PostgreSQL Checkpointer

use agents_sdk::{PostgresCheckpointer, ConfigurableAgentBuilder};
use std::sync::Arc;

let checkpointer = Arc::new(
    PostgresCheckpointer::new("postgresql://user:pass@localhost/agents").await?
);

let agent = ConfigurableAgentBuilder::new("You are a helpful assistant")
    .with_checkpointer(checkpointer)
    .build()?;

§DynamoDB Checkpointer

use agents_sdk::{DynamoDbCheckpointer, ConfigurableAgentBuilder};
use std::sync::Arc;

let checkpointer = Arc::new(
    DynamoDbCheckpointer::new("agent-checkpoints").await?
);

let agent = ConfigurableAgentBuilder::new("You are a helpful assistant")
    .with_checkpointer(checkpointer)
    .build()?;

Modules§

agent
buildertoolkit: Tool builder utilities for creating tools from functions
builtintoolkit: Built-in tools for common agent operations
dynamodb_checkpointeraws: DynamoDB-backed checkpointer implementation for AWS deployments.
events: Event system for agent lifecycle tracking and progress broadcasting
hitl: Human-in-the-Loop (HITL) types for agent execution interrupts.
llm
messaging
persistence: Persistence traits for checkpointing agent state between runs.
prelude: Prelude module for common imports
security: Security utilities for PII protection and data sanitization
state
tools: Core tool system for AI agents

Structs§

AnthropicConfig
AnthropicMessagesModel
ConfigurableAgentBuilder: Builder API to assemble a DeepAgent in a single fluent flow, mirroring the Python create_configurable_agent experience. Prefer this for ergonomic construction.
DeepAgent: Core Deep Agent runtime implementation
DynamoDbCheckpointeraws: DynamoDB-backed checkpointer for serverless AWS deployments.
DynamoDbCheckpointerBuilderaws: Builder for configuring a DynamoDB checkpointer.
EditFileTooltoolkit: Edit file tool - performs string replacement in a file
GeminiChatModel
GeminiConfig
HitlPolicy
LsTooltoolkit: List files tool - shows all files in the agent’s filesystem
OpenAiChatModel
OpenAiConfig
PostgresCheckpointerpostgres: PostgreSQL-backed checkpointer with connection pooling.
ReadFileTooltoolkit: Read file tool - reads the contents of a file
ReadTodosTooltoolkit: Read todos tool - retrieves the current todo list
RedisCheckpointerredis: Redis-backed checkpointer with connection pooling and TTL support.
SubAgentConfig: Configuration for creating and registering a subagent using a simple, Python-like shape.
SummarizationConfig: Configuration for summarization middleware
TokenCosts: Token cost configuration for different providers
TokenTrackingConfig: Configuration for token tracking middleware
TokenTrackingMiddleware: Token tracking middleware that wraps an LLM to monitor usage
TokenUsage
TokenUsageSummary: Summary of token usage across all requests
ToolBuildertoolkit: Builder for creating tools from async functions
ToolContext: Context provided to tool implementations for accessing agent state and utilities
ToolParameterSchema: JSON Schema definition for tool parameters
ToolRegistry: Tool registry for managing and discovering available tools
ToolSchema: Complete schema definition for a tool
UnimplementedSecretsProvideraws: Stub Secrets Manager provider; real implementation will sit behind the secrets feature.
WriteFileTooltoolkit: Write file tool - creates or overwrites a file
WriteTodosTooltoolkit: Write todos tool - updates the agent’s todo list

Enums§

StreamChunk: A chunk of streaming response from the LLM
ToolResult: Result of a tool invocation

Traits§

AgentHandle: Abstraction for hosting a fully configured agent (planner + tools + prompts).
Checkpointeraws: Trait for persisting and retrieving agent state between conversation runs. This mirrors the LangGraph Checkpointer interface used in the Python implementation.
SecretsProvideraws: Placeholder trait for loading configuration secrets.
Tool: Core trait for tool implementations

Functions§

create_async_deep_agent: Async constructor alias to mirror the Python API surface.
create_deep_agent: Create a deep agent - matches Python create_deep_agent() API exactly
create_filesystem_toolstoolkit: Create all filesystem tools and return them as a vec
create_todos_tooltoolkit: Create the todos tool (write only)
create_todos_toolstoolkit: Create both read and write todos tools
get_default_model: Returns the default language model configured Uses OpenAI GPT-4o-mini for cost-effective operation. This model provides excellent performance at a fraction of the cost compared to larger models.
tooltoolkit: Quick helper to create a simple async tool
tool_synctoolkit: Quick helper to create a simple sync tool

Type Aliases§

AgentStream: Type alias for a stream of agent response chunks
ChunkStream: Type alias for a pinned boxed stream of chunks
ThreadIdaws: Unique identifier for a conversation thread/session.
ToolBox: Type alias for boxed tool instances

Attribute Macros§

tooltoolkit: Converts a Rust function into an AI agent tool with automatic schema generation.

Crate agents_sdk

Crate agents_sdk Copy item path

§Rust Deep Agents SDK

§Quick Start

§Features

§Installation Options

§Persistence Examples

§Redis Checkpointer

§PostgreSQL Checkpointer

§DynamoDB Checkpointer

Modules§

Structs§

Enums§

Traits§

Functions§

Type Aliases§

Attribute Macros§

Crate agents_sdk