Back to all categories
Core Sector92 agents curated

Best Agentic Foundations & Multi-Agent Orchestration AI Agents

Multi-agent runtime frameworks, episodic memory vector stores, inference gateways, and agent evaluation suites. Explore our curated selection of top-performing autonomous agents and tooling in this sector, compare pricing and features, and deploy the ideal agent for your stack.

MetaGPT logoFree

MetaGPT

MetaGPT assigns software-team roles to language models. A short requirement becomes plans, documents and code as the agents hand work to one another.

View details Website
xpander.ai logoFree

xpander.ai

xpander.ai brings an AI gateway, an MCP gateway and an agent runtime into an organization's own environment for running and governing agents and coding tools.

View details Website
TinyFish logoPaid

TinyFish

TinyFish supplies serverless web infrastructure for agents that need to navigate live sites, sign in, extract records or take an action in the browser.

View details Website
KaibanJS logoFree

KaibanJS

Each specialist's assignment appears on a Kanban board in KaibanJS. JavaScript developers define the agents' goals and constraints, then follow each task across the board.

View details Website
PraisonAI logoFree

PraisonAI

PraisonAI lets developers run agents on their own computers, including offline setups. Its toolkit covers memory, tools, knowledge and coordination between agents.

View details Website
Together AI logoPaid

Together AI

Together AI hosts open models with serverless or dedicated inference, batch processing and fine-tuning. Teams can also use GPU clusters for post-training.

View details Website
Letta logoPaid

Letta

Memory across sessions is the focus of Letta's research and tooling. Developers use its stateful agents to explore how earlier interactions influence later work.

View details Website
AgentGPT logoFree

AgentGPT

AgentGPT runs in the browser. Give an agent a name and a goal, configure it and start a task such as researching a topic or planning a trip.

View details Website
LangGraph logoPaid

LangGraph

A LangGraph agent can pause for a person's decision and resume with its state intact. Developers define the branches and handoffs of the workflow in an MIT-licensed library.

View details Website
Haystack logoFree

Haystack

Haystack gives developers modular components for retrieval and agent orchestration. Its open-source framework supports building applications that supply models with relevant context.

View details Website
ModelBench logoPaid

ModelBench

Compare the answers different models give to the same prompts in ModelBench, a visual workspace for running benchmarks without code.

View details Website
Cognee logoPaid

Cognee

Graph, vector and relational retrieval work together in Cognee to give agents memory across sessions. The open-source platform runs locally or on Cognee Cloud.

View details Website
Flo AI logoFree

Flo AI

Flo AI now points to Rootflo's Wavefront AI. Its console connects models, agents and tools so data can pass between steps in an application.

View details Website
Groq logoFree

Groq

Groq provides cloud model inference on infrastructure built around its LPU. Developers can use the service to serve model responses at scale.

View details Website
Ollama logoFree

Ollama

Ollama lets developers pull and run open models on their own machines, then call those models from apps or agents. Cloud execution is also available.

View details Website
LangChain logoPaid

LangChain

LangChain and LangGraph provide open-source libraries for building agents. LangSmith adds deployment, traces and evaluations for the resulting applications.

View details Website
Credal logoFree

Credal

Credal lets enterprises build and govern MCP servers over their own tools and information. Employees can query that context from connected assistants.

View details Website
BondAI logoFreemium

BondAI

BondAI provides a command line and server for building agents. Developers can arrange them in a hierarchy, coordinate multiple agents or use reflective designs.

View details Website
AI SDK logoFree

AI SDK

Vercel's AI SDK gives TypeScript developers one open-source toolkit for agents, streaming text, image generation, speech and transcription.

View details Website
LangSmith logoPaid

LangSmith

In LangSmith, a production failure can become a test case for the next agent release. Traces show the model and tool calls behind a failed run, with cost and latency.

View details Website
Dify logoFree

Dify

Dify lets teams build agents and retrieval tools visually, then publish them as an app, API or embed. The same workflow can run on hosted or self-hosted editions.

View details Website
Nexa AI logoFree

Nexa AI

Nexa AI developed on-device inference for language and multimodal models. Its work is now part of Qualcomm AI Hub.

View details Website
Langflow logoFree

Langflow

A diagram built in Langflow can become an API or an MCP server. The low-code workflow canvas connects models, tools and vector stores, with Python for custom components.

View details Website
Browser Use logoFree

Browser Use

Browser Use provides an open-source browser agent and hosted browsers for tasks that need a real session, such as logging in and extracting structured results.

View details Website
Agenta logoPaid

Agenta

Agenta lets teams build an agent in chat, improve it with feedback and use it from Slack or Telegram. The workspace is available in the cloud or self-hosted.

View details Website
agentforge logoFree

agentforge

AgentForge is an open-source framework for composing agent personas, tools and prompt sequences. Developers can connect specialists into a multi-agent system.

View details Website
Firecrawl logoPaid

Firecrawl

One URL can yield Markdown for a model or structured JSON for an app in Firecrawl. Its hosted API and open-source project cover search, crawling and site mapping.

View details Website
Zep logoPaid

Zep

Facts about a customer can change between sessions. Zep stores them in a temporal knowledge graph, giving AI agents long-term memory that tracks those changes.

View details Website
Helicone logoPaid

Helicone

Helicone records model requests as they pass through its AI gateway. Developers can inspect the traffic when debugging or set limits on usage and spending.

View details Website
Agent Herbie logoFree

Agent Herbie

Agent Herbie runs offline on private infrastructure, including air-gapped deployments. Its distributed agents serve operations with limited access to public cloud services.

View details Website
Questflow AI logoFreemium

Questflow AI

Questflow coordinates agents working across applications. It breaks down a task, routes each part to the appropriate agent and runs independent work in parallel.

View details Website
Arize AI logoPaid

Arize AI

Arize offers managed agent observability through Arize AX and self-hosted tracing through the open-source Phoenix project. Both support evaluation and experiments.

View details Website
Mission Squad logoFree

Mission Squad

Mission Squad splits a larger task across specialist agents. It coordinates dependencies, runs independent steps in parallel and collects the results.

View details Website
Linkup logoFree

Linkup

Linkup returns web research with inline citations through a search API that also supports deployment in your own cloud.

View details Website
Rig logoFree

Rig

Rust developers can call several model providers and define agent tools through Rig's type-safe interface.

View details Website
AgentOps logoPaid

AgentOps

Replay a failed AgentOps session to see the sequence of model calls, which tools the agents used and where a handoff between agents went wrong.

View details Website
Restack logoPaid

Restack

Restack records versions of the prompts and tools behind a deployed agent. Its open-source stack also supports experiments and user feedback.

View details Website
Vijil logoFree

Vijil

Vijil tests how agents behave under difficult conditions and applies runtime controls. Teams can use the results when deciding what an agent may do on its own.

View details Website
LangWatch logoFree

LangWatch

LangWatch lets platform teams inspect model traffic, trace agent runs and compare evaluations. They can also simulate scenarios and manage prompt versions.

View details Website
Gumloop logoPaid

Gumloop

Gumloop lets business teams describe an agent's task, pick a model and connect apps such as Slack, HubSpot and Salesforce. IT keeps control of the workflow.

View details Website
Eidolon AI logoFree

Eidolon AI

Eidolon packages an open-source agent server with an SDK. Developers deploy agents as services and call them for tasks such as chat or retrieval.

View details Website
Emergence AI logoFree

Emergence AI

Emergence develops infrastructure for enterprise agents that need to plan actions, coordinate work and verify results under organizational controls.

View details Website
LiteLLM logoFree

LiteLLM

LiteLLM gives applications one OpenAI-compatible interface for multiple model providers. Its open-source gateway routes requests, authenticates callers and caps spending.

View details Website
Pydantic AI logoFree

Pydantic AI

Python types guide both calls to a model and validation of its responses in Pydantic AI. The framework also supports agents, voice, images and embeddings.

View details Website
AgentOS logoFree

AgentOS

AgentOS now points to AG2, an open-source Python framework from the creators of AutoGen for coordinating multiple agents in an application.

View details Website
H Company logoFree

H Company

H Company's computer use agents work across desktop apps, browsers, mobile interfaces and documents to carry out operational tasks.

View details Website
Phala logoPaid

Phala

Phala Cloud runs private model inference, agents and GPU jobs in hardware-backed trusted execution environments.

View details Website
Confident AI logoPaid

Confident AI

Confident AI lets teams compare evaluations with live application traces and expert feedback to track where model or agent quality changes.

View details Website
Voyage AI logoFree

Voyage AI

Embedding a collection and reranking its search results are separate jobs in Voyage AI's model lineup. The MongoDB offering includes general-purpose and domain-specific retrieval models.

View details Website
Flowise logoPaid

Flowise

Flowise is an open-source visual AI agent builder. Its modular canvas can represent a simple model call or a workflow of agents working together.

View details Website
markdown2pdf.ai logoFree

markdown2pdf.ai

Send Markdown with tables, code blocks or LaTeX to markdown2pdf.ai and its API returns a finished PDF. Agents can reach it through MCP and pay per call.

View details Website
Surf.new logoFree

Surf.new

Forms, scrolling and changing page layouts are part of the job for Surf.new. The browser agent uses on-screen controls to carry out web tasks and collect data across pages.

View details Website
Kragent.ai logoFreemium

Kragent.ai

Kragent works through repetitive knowledge tasks in conversation. A professional can revise the instructions as the assistant develops its answer.

View details Website
OpenPipe AI logoFree

OpenPipe AI

OpenPipe provided tools for training language models, evaluating outcomes and inspecting runs. The project migrated to Weights & Biases and CoreWeave.

View details Website
You.com logoFreemium

You.com

You.com offers search, answer and research APIs for developers who want to give an agent current information from the web.

View details Website
Agno logoFree

Agno

Agno runs agents in a team’s own cloud, with controls for memory, knowledge and approval steps. Developers can trace a run and evaluate its results.

View details Website
Guardrails AI logoFree

Guardrails AI

Guardrails AI lets developers define checks on generated outputs and apply those checks across models and deployment environments.

View details Website
StartClaw logoFree

StartClaw

Once deployed on StartClaw, an OpenClaw agent remains available through WhatsApp or Telegram between conversations, without a server for the user to maintain.

View details Website
Browserable logoFree

Browserable

Browserable is an open-source JavaScript library for browser agents. Developers can define navigation, form entry and extraction as part of an application.

View details Website
Anon logoFree

Anon

Anon lets developers record browser actions so an agent can work inside authenticated web applications, including services with limited API access.

View details Website
Inferable logoPaid

Inferable

Inferable keeps track of tool calls, task progress and memory while an agent runs. Developers use its open-source platform to build tasks that can span longer sessions.

View details Website
Coval logoFreemium

Coval

Callers interrupt, switch topics and change booking details. Coval simulates those exchanges so a voice team can inspect the turns where an agent loses track.

View details Website
LlamaCloud logoPaid

LlamaCloud

LlamaCloud, the managed service from LlamaIndex, uses LlamaParse to parse documents. Extracted fields can then be indexed for retrieval.

View details Website
Mirascope logoFree

Mirascope

Mirascope gives Python developers typed model calls and tool definitions. They can follow a request through a trace and inspect its cost.

View details Website
Foundry logoFree

Foundry

Repeatable browser snapshots in Foundry let developers run two web agents against the same page state and compare precisely where they fail.

View details Website
Leverage AI logoFree

Leverage AI

Leverage AI reads purchase orders into manufacturers' and distributors' ERP systems. Buyers can follow shipments and compare supplier performance.

View details Website
ControlFlow logoFree

ControlFlow

ControlFlow lets Python developers define agent tasks as structured steps on Prefect's execution engine, with explicit control over each workflow stage.

View details Website
Langfuse logoPaid

Langfuse

Open a Langfuse trace to see the prompt, tool calls, cost and latency behind an agent response. The open-source platform also manages prompts and compares evaluations across runs.

View details Website
SwarmZero logoPaid

SwarmZero

SwarmZero combines an AI agent marketplace with a visual builder. Specialists can package their expertise as an agent and publish it for other teams.

View details Website
E2B logoPaid

E2B

Code execution, browser tasks and tool calls run in an isolated cloud machine with E2B. Developers can pause or fork the environment in E2B's cloud or their own cloud account.

View details Website
Phoenix logoPaid

Phoenix

Arize Phoenix records traces and evaluates LLM applications in an environment you host. Inspect a run or compare experiments on a dataset.

View details Website
Sema4.ai logoFree

Sema4.ai

Sema4.ai separates agent development, monitoring and human review into distinct spaces. Developers build agents with an SDK while operators handle cases that need human review.

View details Website
Agent Genesis logoFree

Agent Genesis

Agent Genesis publishes MIT-licensed components developers can copy into their own apps. The collection covers document retrieval, chat and agent utilities.

View details Website
CAMEL logoFree

CAMEL

CAMEL is an open-source research framework for studying how agents communicate. Researchers can assign roles, simulate a society of agents or generate data.

View details Website
Pezzo logoFree

Pezzo

Prompts in Pezzo move through development, staging and production with version history. If output deteriorates, roll back the change. Individual traces show token use and cost.

View details Website
TableCharts logoPaid

TableCharts

Give TableCharts a table from Notion, Google Sheets, CSV or Salesforce and it returns a chart URL. The chart can update with the source data and can be passed to another agent.

View details Website
Langroid logoFree

Langroid

Each agent in Langroid is a Python object with its own model, tools and memory. A controller can divide a larger assignment among several of these specialists.

View details Website
Athina AI logoFree

Athina AI

Before shipping a change, compare two agent runs in Athina AI. The same workspace holds the evaluation datasets, annotations and production monitoring.

View details Website
BabyAGI logoFree

BabyAGI

BabyAGI is Yohei Nakajima's research project on self-building agents. Its newer framework stores executable functions and their dependencies in a database, while the original task loop is archived.

View details Website
OpenAGI logoFree

OpenAGI

OpenAGI lets developers assemble an agent from a planner, workers, memory, tools and a vector store. Its open-source examples cover research, writing and travel planning.

View details Website
Upsonic logoFree

Upsonic

An onboarding or document-review agent built with Upsonic can pause at a sensitive step and wait for a person to approve it.

View details Website
Freysa logoFreemium

Freysa

A Freysa personal agent gets its own machine, domain and email address. The owner controls its inbox and web presence.

View details Website
Anchor logoFree

Anchor

Anchor provides cloud computers and browsers for agents that need to use websites. A session can maintain a login across repeated tasks.

View details Website
Agentset logoPaid

Agentset

Agentset answers questions about a private document collection with citations. Its open-source retrieval platform can include tables, graphs and images alongside text.

View details Website
Spell logoPaid

Spell

Spell runs multiple agents on separate tasks in parallel. Watch a run in chat and steer it as it progresses.

View details Website
Tavily logoPaid

Tavily

Ranked passages and their source URLs come back together from Tavily's web search API. Research agents can also extract one article, crawl a site or map its pages before indexing.

View details Website
CrewAI logoFree

CrewAI

Inside a CrewAI crew, one agent can research a subject and hand its findings to another to write the report. Developers assign each agent a role.

View details Website
Self-Operating Computer logoPaid

Self-Operating Computer

Self-Operating Computer lets a model read the screen, click and type. That makes the open computer-use framework relevant to desktop tasks that lack an API.

View details Website
OpenRouter logoFree

OpenRouter

One OpenRouter API key gives an application a choice of model providers. Route requests by model, switch providers on failure and track spending in one place.

View details Website
Blackbox AI logoFree

Blackbox AI

Blackbox offers infrastructure for training and running models on GPUs. Its agents monitor machines and tune training against benchmarks you choose.

View details Website
CorvinOS logoPaid

CorvinOS

Each agent action in CorvinOS leaves a hash-linked audit record. This self-hosted system connects agents to messaging channels and a browser, with consent checks for sensitive steps.

View details Website
Agent Zero logoFree

Agent Zero

Inside Agent Zero's virtual Linux machine, the assistant can write a tool, run it and remember what it learned for the next task. The open-source framework is self-hosted.

View details Website