Conversational assistant
Answers questions and helps reasoning. A conversation alone does not establish external action execution
TECHNOLOGY DIRECTORY · WEBAI
AI AgentsSystems, tools and uses
Explore agents that use tools, research, code or perform Web tasks — and distinguish these products from platforms used to build custom agents. Usage profiles, practical examples, caveats and original documentation
22Documented profiles
12Ready-to-use agents
10Frameworks and platforms
FR / ENBilingual edition
DEFINITIONS
Not every assistant is an agent
A conversational reply, a fixed automation and an agent choosing tools are different mechanisms. Examine agents by their permitted actions, runtime, verifiable outcomes and human oversight
Tool-using agent
Plans and adjusts steps, uses authorized tools and must inspect action outcomes
Agent framework
Developer components to configure models, state, permissions, tools and workflows — not necessarily an immediately usable agent
SOURCE-LINKED DIRECTORY
Find an agent by its role
Profiles prepared from official documentation. Inclusion in WebAI is not a commercial recommendation or independent certification. Availability, features and prices can change
01Ready-to-use agent
ChatGPT Work
OpenAI Work and research
An agentic work environment for multi-step tasks, deliverables and authorized connected apps or files. This is an end-user product, not a framework you embed in your website.
Example use
Produce a source-led comparison dossier and deliver it in a reusable format.
02Ready-to-use agent
ChatGPT Atlas — Agent mode
OpenAI Web navigation
Agent mode in the ChatGPT Atlas browser for navigating pages, clicking and completing Web steps under user control. Available functions depend on browser and security settings.
Example use
Visit three official pages, extract comparable details and draft a sourced report.
03Ready-to-use agent
OpenAI Codex
OpenAI Software engineering
Software development agent for inspecting repositories, preparing changes, running checks and helping review code. Interfaces and permissions vary by environment.
Example use
Fix a documented bug in a branch, add tests and review the diff before merging.
04Ready-to-use agent
Claude Code
Anthropic Software engineering
Coding agent working with files and commands in supported environments. It supports code inspection, editing and testing with permission controls and review.
Example use
Inspect a function, explain the cause of a bug and prepare a patch with a regression test.
05Ready-to-use agent
GitHub Copilot Agents
GitHub Software engineering
Family of coding agents integrated with GitHub workflows and development tools: assisted sessions, delegated tasks and branch or pull-request changes.
Example use
Delegate a small issue and review the resulting pull request before merging.
06Ready-to-use agent
Cursor Agent
Cursor Software engineering
Agentic coding assistant inside the Cursor editor, able to search code, edit multiple files and run project commands.
Example use
Refactor a complex component while preserving behavior and tests.
07Ready-to-use agent
Jules
Google Software engineering
GitHub-connected software agent that creates a plan, works in an isolated development environment and presents proposed changes for review.
Example use
Add tests to a function and prepare a proposed change through GitHub.
08Ready-to-use agent
Gemini CLI
Google Software engineering
Agentic terminal interface for coding and local task automation with Gemini models and configurable tools.
Example use
Inspect project dependencies and prepare an improvement plan without changing production.
09Ready-to-use agent
OpenHands
OpenHands Software engineering
Software-agent ecosystem with Web and command-line experiences, alongside SDKs for developing coding automations.
Example use
Explore a repository, propose a bug fix and gather test results in a controlled environment.
10Ready-to-use agent
Cline
Cline Software engineering
Coding agent integrated with editor workflows, focused on source changes and supervised technical tasks using a configured model.
Example use
Explore a project, fix a layout defect and document changed files.
11Ready-to-use agent
Aider
Aider Software engineering
Terminal coding assistant working with Git for inspectable changes and iterative requests, diffs and tests.
Example use
Change a small function, add a test and inspect the related commit.
12Ready-to-use agent
Manus
Manus Work and research
Agentic service offering research, browsing, document creation and automations depending on mode and subscription. Announced features must be distinguished from account availability.
Example use
Prepare a website comparison report, then separately verify claims and cited sources.
13Agent building framework or platform
Browser Use
Browser Use Web navigation
Library and ecosystem for building browser-controlling agents. Unlike a ready-to-use assistant, the developer configures models, tools, access and execution policy.
Example use
Test an agent navigating a demo catalog and extracting listings without making purchases.
14Agent building framework or platform
LangGraph
LangChain Multi-agent orchestration
Stateful orchestration framework and runtime based on graph nodes and transitions. Useful for agent workflows requiring resumption, interruption and explicit control flow.
Example use
Build a document assistant with retrieval, synthesis and human-approval stages.
15Agent building framework or platform
CrewAI
CrewAI Multi-agent orchestration
Framework for teams of specialized agents (crews) and orchestrated flows. Developers define roles, tasks, interactions and structured outputs.
Example use
Coordinate a source-gathering agent and a drafting agent, followed by human review.
16Agent building framework or platform
OpenAI Agents SDK
OpenAI SDKs and frameworks
SDK for composing agents with tools, specialist handoffs, guardrails, sessions and tracing. A developer runtime rather than a finished standalone agent product.
Example use
Create an agent that triages requests, calls a dummy function and produces structured output.
17Agent building framework or platform
Google Agent Development Kit
Google SDKs and frameworks
Development framework for agents and multi-agent systems combining models, tools and testing. Provides building blocks instead of a ready-made directory of agents.
Example use
Create two agents for retrieving documents and summarizing results with references.
18Agent building framework or platform
Mastra
Mastra SDKs and frameworks
TypeScript framework for agents, tools and workflows with memory, integrations and observability according to configured modules.
Example use
Build document retrieval with constrained tools and an auditable summary output.
19Agent building framework or platform
PydanticAI
Pydantic SDKs and frameworks
Python library for agent applications with data validation, typed dependencies and structured tools. Useful where inputs and outputs must follow explicit schemas.
Example use
Extract bibliographic metadata into a validated structure and detect missing fields.
20Agent building framework or platform
n8n — AI Agent node
n8n Visual workflows
Visual automation component adding model-driven tool use to n8n workflows. Adding an AI node does not make every step agentic.
Example use
Build a source-monitoring, classification and proposed-notification workflow awaiting approval.
21Agent building framework or platform
Dify
Dify Visual workflows
Platform for building AI apps and workflows using models, tools, document sources and processing logic. Features vary by edition and integration.
Example use
Create a documentary research assistant using a small public corpus with citations to review.
22Agent building framework or platform
Flowise
FlowiseAI Visual workflows
Visual environment for assembling model applications and agents using tools and retrieval flows. Execution graphs require review before using sensitive data.
Example use
Prototype an agent that reads public documents, extracts passages and drafts a sourced answer.
No profiles match these filters. Change your search or reset the criteria
PRACTICAL METHOD
How to use or build an agent
Method depends on environment and permissions: a Web assistant, a coding agent and a developer SDK are set up and evaluated differently
Select a ready-to-use agent
Start with the outcome rather than marketing language: sourced research, coding, browsing or deliverables. A conversational assistant is not necessarily an agent that executes actions.
- Describe a reversible task and its verifiable outcome
- Confirm actual mode availability and access terms
- Limit permitted documents, accounts and permissions
- Monitor steps and reject unauthorized actions
- Check the final result and retain evidence
Build an agent for the Web
A custom agent combines instructions, a model, tools and authorization policy. A framework cannot guarantee safe tool use or factual accuracy: those must be designed and evaluated.
- Choose one job and break it into operations
- Define tool inputs, outputs and schemas
- Limit permissions and isolate untrusted external data
- Require approval before consequential actions
- Measure failures, costs, reproducibility and no-answer cases
SAFETY AND RELIABILITY
Delegating a task does not delegate responsibility
Web pages, files and tool outputs remain potentially untrusted data. Credentials, commands, sent messages, payments and deletions need appropriate oversight. Success on one scenario does not certify general reliability
01Limit access and isolate the runtime
02Require approval before consequential actions
03Record sources, operations and failures
04Check the final state rather than trusting a success message
FAQ
Common questions about AI agents
01How does an AI agent differ from a chatbot?
A chatbot primarily converses and answers. An agent pursues a multi-step goal, calls tools and may inspect action outcomes. Products may combine both experiences.
Read the source ↗02Is an agent-building platform a ready-to-use agent?
Not necessarily. LangGraph, CrewAI and SDKs provide development frameworks. Models, tools, permissions, infrastructure and evaluation still need configuration.
Read the source ↗03Can agents be used for free?
It depends on the product, hosting and model. Some are open source or offer free tiers, but inference, integrations, usage quotas and hosting may be billed. Always check current official terms.
Read the source ↗04Can an agent safely operate my accounts without risk?
No. Web, tool and account access introduce risks: unintended actions, malicious page content, data leakage and errors. Limit permissions and require approval for consequential operations.
Read the source ↗05Are MCP, A2A and AI agents the same thing?
No. An agent carries out tasks using tools; MCP defines exchanges with tools and resources; A2A addresses communication between agents. Protocols alone do not ensure safety or authorization.
Read the source ↗FOUNDATIONAL DOCUMENTATION
Original sources for independent verification
These references explain agents, tools and risks. They do not endorse the 22 listings or constitute sponsored rankings
GitHub Docs — About agentsOpenAI — Agents SDKAnthropic — Claude CodeGoogle — Agent Development KitOWASP — Prompt injectionMCP — Protocol documentation
Editorial status: pending independent review. Profiles draw on official documentation and need periodic updates. No agent is directly connected to this portal