Agentic Tool Calling With Multi Step Reasoning And State Management

via “agentic reasoning with iterative tool invocation and state management”

Production NLP/LLM framework for search and RAG pipelines with component-based architecture.

Unique: Implements agents as composable pipeline components with explicit state management and tool registry, supporting both synchronous and asynchronous execution — combined with schema-based tool definition that automatically converts to provider-specific formats (OpenAI function_call, Anthropic tool_use) without manual serialization

vs others: More transparent than LangChain's AgentExecutor (which abstracts the reasoning loop) and more flexible than AutoGPT (which is a fixed architecture) — allowing custom agent implementations while providing production-ready defaults

via “agent loop execution with tool-use reasoning and step-by-step planning”

Drag-and-drop LLM flow builder — visual node editor for chains, agents, and RAG with API generation.

Unique: Implements a generalized agent loop that supports multiple reasoning patterns (ReAct, Plan-and-Execute) through configurable LLM prompts and tool schemas. The system tracks agent state across iterations, enforces step limits, and logs each reasoning step for observability and debugging.

vs others: More transparent than black-box agent frameworks because step-by-step reasoning is logged and inspectable; more flexible than single-pattern agents because reasoning strategy is configurable via prompts.

via “agentic tool calling with multi-step reasoning and state management”

The AI Toolkit for TypeScript. From the creators of Next.js, the AI SDK is a free open-source library for building AI-powered applications and agents

Unique: Implements a provider-agnostic agentic loop that normalizes function calling across OpenAI, Anthropic, Google, and other providers. Uses a unified tool schema format (Zod-based) that's converted to provider-specific formats at runtime. Supports middleware-based tool execution, allowing custom logging, error handling, or result transformation without modifying core agent logic.

vs others: Simpler than LangChain's AgentExecutor (no complex state management classes) and more flexible than provider-specific SDKs, with built-in support for streaming tool results and middleware-based extensibility.

via “agentic planning and multi-step execution”

Google's multimodal API — Gemini 2.5 Pro/Flash, 1M context, video understanding, grounding.

Unique: Supports agentic planning where the model decomposes tasks into steps and decides which tools to call, with the client orchestrating the execution loop, enabling flexible multi-step workflows without hardcoded task logic

vs others: More flexible than pre-defined workflow systems because the model decides the execution plan, but requires more client-side orchestration logic than fully managed agent platforms like Anthropic's Claude with tool use

via “agentic reasoning loop with tool-use planning”

an open source, extensible AI agent that goes beyond code suggestions - install, execute, edit, and test with any LLM

Unique: Implements a stateful reasoning loop that maintains execution context across iterations, with explicit state tracking (thinking → tool-calling → observing → deciding) rather than a simple request-response pattern. Supports both synchronous and asynchronous execution modes, allowing agents to schedule long-running tasks and return to the user.

vs others: More sophisticated than simple tool-calling because it includes planning and reasoning steps; more practical than pure LLM agents because it integrates real tool execution and observes actual results rather than simulated outputs.

via “chain-of-thought reasoning within function-calling loop”

Latest compact reasoning model with native tool use.

Unique: Reasoning loop is native to the model's forward pass rather than a post-hoc wrapper; the model's internal computation directly influences tool selection and parameter refinement, not just the final response. This differs from frameworks that apply reasoning as a separate preprocessing step before tool calling.

vs others: Tighter integration of reasoning and tool use than GPT-4o or Claude 3.5 Sonnet, which treat reasoning and function calling as sequential stages; o4-mini's interleaved approach reduces hallucinated tool parameters and improves error recovery in multi-step workflows.

via “agent framework with multi-step reasoning and tool integration”

Unified framework for building enterprise RAG pipelines with small, specialized models

Unique: Integrates agentic reasoning (ReAct pattern) with llmware's retrieval and small model ecosystem, enabling cost-effective multi-step workflows. Supports both agentic loops (non-deterministic) and DAG-based workflows (deterministic) for different compliance requirements. Tool integration is flexible, supporting custom APIs and code execution.

vs others: Integrated with llmware's small model ecosystem for cost-effective multi-step reasoning vs LangChain agents using large LLMs; supports both agentic and deterministic workflows vs pure agentic frameworks; built-in retrieval integration vs external RAG systems.

via “react agent-driven reasoning with tool orchestration”

Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.

Unique: Combines ReAct reasoning with dependency-injected tool orchestration and multi-turn session management, allowing agents to reason across heterogeneous data sources (KB, web, MCP tools) while maintaining conversation context. Supports both streaming and batch reasoning modes.

vs others: More transparent and debuggable than black-box agent frameworks (reasoning steps are visible), more flexible than fixed RAG pipelines (can adapt strategy per query), and more cost-efficient than multi-turn LLM calls by batching reasoning and retrieval.

via “agentic multi-step reasoning with tool integration”

SoTA production-ready AI retrieval system. Agentic Retrieval-Augmented Generation (RAG) with a RESTful API.

Unique: Combines local RAG retrieval with web search in a single agent loop, enabling fallback to external sources when knowledge base lacks information. Streaming responses expose intermediate reasoning steps, allowing clients to display agent thinking in real-time. Tool schema registry is provider-agnostic, supporting OpenAI, Anthropic, and custom LLM backends.

vs others: More transparent than LangChain agents because streaming exposes all reasoning steps; more flexible than Vercel AI's tool calling because it supports local LLM backends (Ollama) without cloud dependency.

via “agentic reasoning with multi-step task decomposition”

runs anywhere. uses anything

Unique: Implements explicit state transitions between planning, execution, and reflection phases, where each phase produces structured artifacts that are fed back into the reasoning loop, enabling agents to learn from failures and adapt plans rather than just executing a static sequence

vs others: More transparent than black-box agent frameworks because reasoning steps are visible and auditable; more robust than single-shot approaches because agents can recover from failures through reflection

via “agent-based reasoning and tool orchestration”

A data framework for building LLM applications over external data.

Unique: Provides a unified Agent abstraction supporting multiple reasoning architectures (ReAct, function-calling, custom) with automatic tool binding and execution tracing. Tools are defined declaratively with schema and implementation, enabling agents to discover and use them without manual integration code.

vs others: More flexible agent architecture than LangChain's agents; better execution tracing and debugging support for complex multi-step reasoning.

via “agent orchestration with multi-step reasoning and tool loops”

The LLM Anti-Framework

Unique: Implements agent loops as a first-class abstraction with built-in support for tool calling, result processing, and conversation history management. Unlike LangChain's AgentExecutor (which requires custom tool definitions and action schemas), Mirascope agents use the same tool system as regular function calls, reducing boilerplate.

vs others: Simpler agent setup than LangChain (reuses tool definitions) and more flexible than AutoGPT-style agents (supports multiple providers and custom stopping conditions), while maintaining Mirascope's provider-agnostic approach.

via “multi-step agentic reasoning with loop control”

We’ve been working with automating coding agents in sandboxes as of late. It’s bewildering how poorly standardized and difficult to use each agent varies between each other.We open-sourced the Sandbox Agent SDK based on tools we built internally to solve 3 problems:1. Universal agent API: interact w

Unique: Provides a pluggable reasoning strategy system where developers can inject custom logic at each step (pre-LLM, post-LLM, tool execution) without modifying the core loop, enabling experimentation with novel reasoning patterns

vs others: More flexible than Langchain's agent executors because it exposes reasoning hooks at finer granularity, allowing custom strategies like tree-of-thought or beam search without forking the framework

via “multi-turn agentic reasoning with document context”

Hi HN,I built an open-source AI agent that has already indexed and can search the entire Epstein files, roughly 100M words of publicly released documents.The goal was simple: make a large, messy corpus of PDFs and text files immediately searchable in a precise way, without relying on keyword search

Unique: Implements agentic reasoning specifically for document investigation, likely with custom tool definitions for search, retrieval, and entity extraction tailored to investigative workflows

vs others: More powerful than single-turn Q&A because the agent can refine searches and reason over multiple documents, but requires more careful prompt engineering to avoid hallucination and inefficient reasoning paths

via “multi-step reasoning with tool invocation across conversation turns”

‘It took nine seconds’: Claude AI agent deletes company’s entire database

Unique: Claude's extended context window and stateful conversation model allow the agent to retain full conversation history including tool results, enabling it to reason about complex workflows without explicit state management or workflow definition files — the agent infers the workflow from the conversation

vs others: More flexible than rigid workflow engines (e.g., Apache Airflow) because the agent can adapt its approach based on results, but less predictable because the reasoning process is not explicitly defined and can vary based on model behavior

via “agentic reasoning with tool calling and multi-step planning”

Build AI Agents, Visually

Unique: Implements agent execution via a dedicated Agentflow execution engine (Agentflow Execution section in DeepWiki) that separates agent logic from chatflow logic; agents use a schema-based function registry that maps tool definitions to LLM function-calling APIs, and the system tracks tool call history and reasoning steps for observability and debugging

vs others: More flexible than LangChain's built-in agents because Flowise allows users to define custom tools and stopping conditions via the UI, and the execution engine provides detailed logging of agent reasoning without code changes

via “agent-based task decomposition with tool calling”

LLM framework to build customizable, production-ready LLM applications. Connect components (models, vector DBs, file converters) to pipelines or agents that can interact with your data.

Unique: Implements agentic loop with schema-based tool registration supporting both function-calling APIs (OpenAI, Anthropic) and ReAct prompting, with automatic tool execution and conversation history management — enabling multi-step reasoning without manual orchestration

vs others: More integrated with RAG pipelines than LangChain agents; better tool schema validation than raw function-calling APIs

via “agentic-workflow-orchestration”

A lightweight agentic workflow system for testing AI agent flows with local LLMs and tool integrations

Unique: Implements a simple but explicit agent loop pattern (think → act → observe) optimized for testing and debugging rather than production scale, with built-in logging for each reasoning step

vs others: Simpler and more transparent than frameworks like AutoGPT or BabyAGI for understanding agent behavior; trades production features (persistence, distribution) for clarity and ease of modification

via “iterative agent reasoning with step-by-step execution”

Hey HN! We launched a thing today, and built a cool demo that I'm excited to share with the community.This tool creates AI agents easily and can handle some really technically complex work. I whipped up this rocket scientist agent in our tool in 10 minutes. I asked a couple of aerospace enginee

Unique: Provides visual step-by-step execution traces within the agent composition interface, making reasoning transparent to non-technical users and enabling iterative refinement based on observed reasoning quality

vs others: Offers better visibility into agent reasoning than black-box API calls, enabling domain experts to validate correctness and iterate on agent behavior without requiring ML expertise

via “multi-step reasoning with internal thought chains”

Proactive personal AI agent with no limits

Unique: Maintains explicit reasoning state across steps with backtracking capability, allowing the agent to revise earlier conclusions rather than committing to single-pass inference like most LLM-based agents

vs others: Provides better explainability than black-box agents by exposing intermediate reasoning, though at the cost of increased latency compared to single-pass inference approaches