Local FirstSelf Hosted
OpenClaw is a local-first personal AI agent that can work across messaging apps, browser tasks, files, and system tools from a self-hosted setup.
Very ambitious local-first agent architecture
Strong tool and memory story
Verified AI Agents Collection
AI Agents
MCP-style tool and SaaS integrations
Through March and April 2026 the agent category matured from “cool demos” into **production harnesses**: coding agents that loop on repos with tests and PRs, **browser agents** that extract data and drive legacy portals when APIs do not exist, and **workplace copilots** wired into Slack, GitHub, Jira, CRMs, and data warehouses—often through **Model Context Protocol (MCP)** servers and IDE agent plugins. Teams pair graph orchestration (LangGraph, n8n, Dify) with guardrails, human handoff, and observability. We map frameworks, managed platforms, and local-first assistants so you can align architecture with latency, cost, and compliance.
Featured AI Agents
Verified AI Directory
Browse our complete database of tested and ranked AI applications.
Self HostedTelegram
Hermes Agent is Nous Research's open-source, self-hosted personal agent with a learning loop, SQLite-backed memory, MCP extensibility, and gateways for Telegram, Discord, Slack, WhatsApp, Signal, and CLI.
Strong grassroots momentum as a full personal-agent stack
Model-agnostic providers including OpenRouter and Ollama
Autonomous CodingSoftware Engineering
Devin is Cognition's autonomous software engineering agent that plans, writes code, runs tests, and iterates in a dedicated environment for end-to-end development tasks.
Built for full-task coding workflows rather than single completions
Strong positioning for teams delegating scoped engineering work
Coding AgentOpen Source
OpenHands is an open platform for coding agents that can inspect codebases, run commands, debug issues, and execute software tasks in controlled environments.
One of the strongest open coding-agent projects
Good for real repo inspection and debugging
LangGraph is a graph-based orchestration framework for building stateful, long-running AI agents with retries, branching, and human-in-the-loop control.
CrewAI is a popular framework for building multi-agent systems where specialized agents collaborate on complex business and automation workflows.
OpenAI Agents SDK is a lightweight framework for building tool-using and multi-agent workflows with handoffs, tracing, and guardrails.
Browser AutomationWeb Agents
Browser Use is an open-source Python layer that connects LLMs to real browser sessions so agents can navigate, extract data, and complete multi-step web tasks—often paired with orchestrators like n8n or frameworks for production web agents in 2026.
Popular default for custom web agents when sites lack APIs
Model-agnostic and easy to combine with workflow tools
Browser AutomationForms
Skyvern is an open-source computer-vision browser agent for automating form-heavy and legacy web workflows—insurance, government, and procurement portals—with natural-language goals instead of brittle selectors alone.
Strong fit when workflows are UI-heavy and APIs are missing
Aligns with 2026 demand for operational browser agents
Web ScrapingCrawl
Firecrawl provides crawl, scrape, and search APIs many teams use as the web data layer for research agents, monitoring bots, and RAG pipelines—feeding clean markdown or structured output into downstream LLM agents.
Common building block in 2026 agent stacks for grounded web context
Speeds up research and competitive intel agents
Browser AutomationSDK
Stagehand is a browser agent SDK that combines natural language instructions with code-level control for reliable web automation.
One of the best browser stacks for agent reliability
Balances natural language with deterministic code
PydanticAI is a Python framework for building production AI agents and workflows with strong typing, validation, and provider flexibility.
TypeScriptFramework
Mastra is a TypeScript framework for building AI agents, workflows, MCP servers, and production-ready AI applications.
Strong TypeScript-first agent framework
Covers workflows and MCP use cases well
Dify is a production-focused platform for building agentic workflows, AI apps, and RAG systems with a visual interface and self-hosting options.
Semantic Kernel is Microsoft's SDK for building AI agents and production AI applications with tools, memory, connectors, and workflow support.
Python.NET
Microsoft Agent Framework is a Python and .NET framework for orchestrating agents, workflows, and production-grade multi-agent systems.
Enterprise-friendly architecture
Good fit for Microsoft-heavy teams
AI IDECascade
Windsurf is Codeium's AI-native IDE with Cascade agentic flows for multi-step coding, refactors, and deep codebase work alongside traditional editing.
Strong agentic IDE experience with flow-state workflows
Competitive alternative to other AI-first editors
GeminiAgent Manager
Google Antigravity is an agent-first development environment built on a VS Code foundation, with Gemini-powered agents, browser-in-the-loop automation, and multi-agent orchestration.
Clear agent-first UX with orchestration primitives
Browser-in-the-loop support for realistic web tasks
AWSSpec Driven
Kiro is AWS's agentic development environment for spec-driven engineering, background agent hooks, terminal workflows, and autonomous tasks that integrate with modern repos.
Spec-driven workflows help reduce ambiguous agent work
Agent hooks and background automation fit real SDLC patterns
IDE PluginZenflow
Zencoder is an agentic coding platform with IDE plugins and Zenflow orchestration for spec-driven workflows, multi-repo context, parallel agents, and verification loops.
Multi-model verification story for higher-trust outputs
Multi-repo indexing fits large org codebases
EnterpriseMicrosoft 365
Microsoft Copilot Studio is a low-code environment for building copilots and agents that connect to Microsoft 365, Dataverse, and enterprise systems with policy controls.
Excellent fit for Microsoft-centric organizations
Enterprise governance and connector ecosystem
OpenAIBrowser Agent
ChatGPT agent mode (evolved from Operator) lets ChatGPT use tools like a virtual browser to take multi-step actions, research, and produce deliverables inside ChatGPT.
Huge reach for users already in ChatGPT
Combines browsing-style action with strong models
AnthropicDesktop
Claude Cowork brings Anthropic's agentic architecture to Claude Desktop for knowledge work: local files, multi-step tasks, and deliverables beyond a single chat turn.
Strong fit for non-developer power users who live in documents
Agentic execution with desktop context
Open SourceWorkflows
n8n is an open workflow automation platform with AI nodes, community templates, and self-hosting options for building agent-like automations across APIs and data sources.
Excellent for engineers who want control and self-hosting
Huge integration surface via community nodes
SaaSAutomation
Zapier Agents lets teams create AI teammates connected to business systems so agents can execute work across thousands of SaaS apps with monitoring and templates.
Massive app graph out of the box
Fast time to value for business operators
AutomationVisual Builder
Make (formerly Integromat) provides visual automation with AI steps and agent-style scenarios for connecting apps, transforming data, and running monitored workflows.
Mature scenario engine with strong connector breadth
Good middle ground between Zapier-style ops and code
Google CloudGemini
Vertex AI Agent Builder is Google's managed toolkit for building, evaluating, and deploying agents grounded in enterprise data with Gemini and Google Cloud security controls.
Strong fit for regulated teams on Google Cloud
Grounding and evaluation hooks for production agents
VS CodeOpen Source
Cline is a VS Code extension coding agent that can read repositories, propose multi-file edits, run commands, and iterate with tool use for software tasks.
Popular open path for agentic coding inside VS Code
Strong community iteration speed
TerminalGit
Aider is a terminal-based AI pair programmer that edits git repositories with multi-file changes, tests, and commit-oriented workflows for agentic coding in the CLI.
Excellent for developers who want reviewable commits
Works with many models and local setups
SuperAGI is a dev-first autonomous agent framework for spawning, managing, and extending multiple agents with tool integrations.
AI AgentsOpen Source
Autogen is a open source framework for building AI agents in Python and .NET.
Live Interface Preview
AI AgentsOpen Source
Eliza is an open source conversational AI agent for Twitter, Discord, and other platforms through integrations.
Live Interface Preview
GeminiFramework
Google's Agent Development Kit is a code-first framework for building, evaluating, and deploying sophisticated agents, especially around Gemini and agent-to-agent workflows.
Good fit for Google-stack agent builders
Built with evaluation and deployment in mind
Agent Zero is a general-purpose AI agent framework that can use the terminal, browser, files, and reusable skills to complete autonomous tasks.
Open SourceGeneral Agent
OpenManus is an open-source general agent framework inspired by Manus, designed for multi-step task execution, tooling, and autonomous workflows.
Popular open-source alternative to Manus
Good for developers prototyping general agent workflows
SWE-agent is a software engineering agent built by researchers at Princeton and Stanford for fixing GitHub issues and handling developer-oriented tasks.
smolagents is Hugging Face's lightweight library for code-first agents with minimal abstractions, tool use, and model-agnostic execution.
Letta is a platform for building stateful agents with advanced memory, self-improving behavior, and persistent long-term context.
MetaGPT is a multi-agent framework that models an AI software company with distinct roles like product manager, architect, and engineer.
CAMEL-AI is a multi-agent framework and research ecosystem focused on scalable agent collaboration, simulation, and automation.
Open SourceDesktop App
Goose is a native open-source AI agent with desktop, CLI, and API interfaces for coding, automation, research, and workflows.
Strong local-first agent experience
Works across desktop, CLI, and API surfaces
MolmoWeb is an open multimodal web agent from Ai2 that uses visual understanding to control a browser and complete online tasks.
OpAgent is a web navigation agent framework built for browser automation and multi-step online task execution.
GenericAgent is a minimal self-evolving agent framework with browser, terminal, filesystem, and device control capabilities.
AutoGPT is one of the best-known autonomous agent projects for building, deploying, and managing long-running AI agents.
Flowise is a visual builder for AI agents, workflows, and multi-agent applications with self-hosting and cloud options.
AgentScope is a developer-centric framework for building observable, production-ready, multi-agent and multimodal AI systems.
Haystack is an orchestration framework for production AI systems that now supports agent workflows, retrieval pipelines, and explicit control over state and tools.
No CodeBusiness Automation
Lindy is a no-code agent builder for business workflows that connects email, calendars, CRMs, and web actions so non-developers can deploy agents with guardrails.
Fast path for operators who want agents without engineering
Strong template and integration mindset
No CodeAutomation
Gumloop is a visual AI automation platform for building agents and workflows across SaaS tools with drag-and-drop orchestration and monitoring.
Approachable for teams that think in workflows
Useful for marketing, ops, and growth automations
No CodeEnterprise
MindStudio is a visual platform to design, test, and deploy AI agents with hundreds of model and integration options for teams and enterprises.
Large integration surface for real business systems
Model-agnostic routing is useful for procurement-heavy teams
EnterpriseCRM
Salesforce Agentforce is Salesforce's enterprise agent layer for CRM-grounded assistants and autonomous actions across Sales, Service, and industry clouds.
Deep native tie-in to Salesforce objects and workflows
Strong story for regulated sales and service operations
BrowserAutonomous
AgentGPT is a browser-based platform for configuring autonomous AI agents with goals, tool access, and iterative execution for experimentation and lightweight automation.
Easy on-ramp for trying autonomous loops
Good learning tool for agent concepts
General AgentBrowser Operator
Manus is a general AI agent product built for multi-step task execution across web research, planning, and complex online workflows.
Strong autonomous task framing and execution
Browser operator expands real-world task coverage
Visual BuilderSelf Hosted
Langflow is an open-source visual builder for AI agents, RAG pipelines, and MCP-backed workflows with APIs for embedding flows into your own apps.
Huge community and template ecosystem for experimentation
Good bridge between no-code flows and production APIs
Self HostedMessaging
LettaBot is an Apache-licensed personal assistant from the Letta team that unifies Telegram, Slack, Discord, WhatsApp, and Signal with persistent memory and local tool execution powered by the Letta stack.
Hackable TypeScript codebase aimed at agent builders
One agent memory across many chat surfaces
TelegramSelf Hosted
Chibi is a MIT-licensed, self-hosted Telegram bot that orchestrates multiple LLM providers, MCP tools, async jobs, and sub-agents from a single mobile-first interface.
Purpose-built for Telegram-first power users
Lightweight enough for small VPS or Raspberry Pi class hardware
Self HostedOllama
Open WebUI is a self-hostable web interface for local and remote LLMs with pipelines, tool use, RAG, and community extensions that many teams treat as a private ChatGPT plus agent playground.
One of the most starred self-hosted AI UIs with active development
Great on-ramp for local models plus cloud fallbacks
Self HostedPlugins
LibreChat is an open-source, self-hosted ChatGPT-style UI with plugins, agents, multimodal chat, and enterprise-friendly authentication patterns.
Mature self-hosted alternative with a serious feature matrix
Strong fit for teams that want OpenAI-compatible routing plus control
Self HostedRAG
AnythingLLM by Mintplex Labs is a self-hosted all-in-one desktop and Docker app for RAG, agents, and multi-workspace document chat over your own data.
Friendly packaging for teams that want private document agents
Supports many model providers and vector setups
Self HostedNotes
Khoj is an open-source personal AI copilot for your notes, docs, and second brain with self-hosting, research agents, and assistant integrations.
Beloved by PKM and research communities for local-first use
Strong angle on grounded answers over your own corpus
VS CodeOpen Source
Roo Code is a community-driven VS Code extension fork in the Cline lineage focused on autonomous coding, deep tool use, and rapid iteration from the open agent ecosystem.
Grassroots velocity: frequent releases and loud community feedback loops
Familiar UX if you already like Cline-style agents
Open Source AIAI Agents
phidata is a framework and interface for building AI agents with 'memory, knowledge, tools and reasoning'.
Live Interface Preview
Expert Research Tips
Verified for 2026
Model Depth & Logic
Higher parameter counts (70B+) directly correlate with better logic and memory persistence in AI Agents.
Privacy & Encryption
Prioritize platforms with End-to-End Encryption or strict "No-Log" policies for sensitive creative sessions.
Our research team monitors API updates and model releases daily to ensure these technical insights remain accurate.
Updated: Apr 2026
Complete 2026 Guide
AI agents after the March–April 2026 wave: how people build today
What counts as an AI agent here?
We treat AI agents as systems that plan, call tools, and iterate until a task completes—not single-shot chat. Typical surfaces today include IDE and terminal agents (multi-file edits, shells, CI), browser runtimes (Playwright/CDP, remote browsers, “computer use” style loops), workflow orchestrators (nodes + LLM steps + webhooks), desktop and DM gateways (files, notifications, chat channels), and frameworks you host yourself. Pure chat without durable tool use stays in chat categories.
What teams actually ship in early 2026
- 1.Coding and shipping loops — agents open issues, propose patches, run tests, and open PRs; humans review. The focus moved from autocomplete to long-horizon tasks with auditable steps.
- 2.Browser and “systems of record” glue — pricing monitors, internal admin portals, insurance or HR forms, and legacy SaaS flows where brittle selectors fail; teams mix LLM navigation with scripted steps for cost and reliability.
- 3.MCP-backed internal tools — connect models to GitHub, ticketing, docs, and cloud APIs through shared tool servers instead of one-off custom clients per vendor.
- 4.Personal and ops agents — always-on assistants on Telegram/Slack or local gateways that summarize alerts, triage inbox-style work, and run scheduled checks.
- 5.Governed enterprise rollouts — SSO, allowlists, secret scanning on agent edits, retention policies, and human-in-the-loop approvals before outbound messages or purchases.
MCP, plugins, and the integration layer
Model Context Protocol (MCP) has become the default pattern for exposing tools, prompts, and resources to agents across Claude, VS Code–family editors, Cursor-class IDEs, and many frameworks. In practice that means fewer bespoke integrations per model vendor and more shared servers for GitHub, databases, ticketing, and internal APIs. It also means new engineering work: tool curation, least-privilege scopes, tracing, and managing context spent on tool metadata in large sessions. The public spec and ecosystem live at modelcontextprotocol.io.
Up-to-date aggregate sources (buyer guides and directories)
Independent roundups help with pricing snapshots and category coverage; shipping details still change weekly—verify on the vendor site before procurement.
- 1.AI Agent Square — agent directory and comparisons (broad enterprise and dev categories, 2026 refresh).
- 2.Fello AI — best AI agents tested roundup (workflow vs coding vs enterprise breakdown).
- 3.ToolCenter — best AI agents 2026 ranked (coding and framework emphasis).
- 4.AI Agent Square — 2026 buyer guide (category winners and evaluation framing).
- 5.AI for Code — AI coding agents directory (developer-focused rankings).
- 6.Faros — AI coding agents 2026 (community and practitioner angles).
- 7.DEV — top agent frameworks 2026 (LangGraph, CrewAI, vendor SDKs, and tradeoffs).
- 8.DEV — MCP in 2026 (protocol, production patterns, ecosystem).
How to pick without regret
Match the product class to the risk surface: coding and browser agents need session isolation, logging, and rollback; customer-facing agents need brand-safe prompts and escalation paths. Budget for tokens plus infra (hosted browsers, queues, GPU workers), not only per-seat SaaS. Prefer stacks with evals, tracing, and replay over opaque magic. For browser-heavy work, plan hybrid automation: deterministic steps where the DOM is stable, LLM steps where it is not. For coding, insist on CI feedback in the loop and human review before merge.
Related Categories
Open source AI
Freely available AI technologies and platforms that encourage collaboration and innovation.
Chat Bots
The best AI chat bots and models for text and voice.
Coding
AI tools to help with programming, code generation, and software development.
AI Search
AI-powered search engines and tools for information retrieval.
Research
The study and development of new AI technologies and methodologies.
View this category list →