AgentenCode / Intelligence Layer
AI ecosystems: Which agents do OpenAI, Google, Anthropic and others offer?
Every agent remains visible in the full directory. Here you can see which separate products belong to which provider or product ecosystem, and where access, scope and safety evidence differ.
AI ecosystems
AGENTENCODE RESEARCH
Multi-Agent Intelligence: Research & Architecture
How collaborating AI agents work: orchestration, delegation, external benchmarks and auditable evidence.
Explore the research guide → · Orchestration patterns · Studies & benchmarks · Evidence methodology
EXPLORE
Seven AI ecosystems at a glance
Select a provider to inspect its documented agents and product layers in context.
ChatGPT & Codex
Research, work and coding agents across ChatGPT and distinct OpenAI developer surfaces.
Claude
Claude Code for development, Cowork for computer and file tasks, and Claude Research for investigation.
Gemini & Entwickleragenten
Gemini research, Gemini CLI, Jules, Spark and the Antigravity developer ecosystem.
Gemini Enterprise Agent Platform
Gemini work agent, Gemini Enterprise Agent Platform and specialized algorithm agent AlphaEvolve.
Copilot & Foundry
Research, analysis, security and development agents across Microsoft 365, Copilot Studio and Foundry.
Computer, Comet & Research
Deep Research, Computer and Comet Assistant span research and varying degrees of delegated task execution.
Muse
Muse as a personal task agent with its own VM and documented action approval mechanisms.
Agent, platform and model are not interchangeable.
A reliable answer must distinguish the right product layer. Models generate outputs; agents execute tasks under defined permissions; agent platforms provide building, deployment and governance capabilities. A model name therefore does not prove a particular agent feature.
AgentenCode curates product associations without duplicating profile identifiers. These relationships do not transfer field-level security evidence. An undocumented property stays Unknown, never automatically “No”.
AGENTENCODE / ECOSYSTEM GUIDE
What makes an AI agent ecosystem?
An AI agent ecosystem is not a single model. It connects distinct agents, development tools, runtimes, permissions and provider products. A defensible assessment compares concrete deployed products—not just the logo or latest model family. Choosing an ecosystem requires separating the provider relationship from product capabilities. A tool connected to ChatGPT does not become identical to Codex; an agent using Gemini models does not automatically inherit Gemini Enterprise security controls. The same caution applies to API frameworks versus consumer applications.
| Layer | How to read it |
|---|---|
| Model | Predicts or generates content; model availability alone says nothing about external action permissions. |
| Agent | Uses context, tools and a specific runtime to plan or execute a task within a defined scope. |
| Framework | Developer components for assembling agent workflows; not itself proof of an end-user agent. |
| Agent platform | Infrastructure for deployment, orchestration, identity and monitoring. Controls vary by product and plan. |
| Ecosystem | A network of provider products, developer surfaces and agents with documented relationships. |
Seven providers, different product layers
All 130 AgentenCode entries remain in one directory. The ecosystem grouping below includes only explicitly linked profiles; it is not a rating or a market-wide census.
| Ecosystem | Catalogue scope | Examples | Source |
|---|---|---|---|
| OpenAI | 6 profiles | ChatGPT Deep Research, OpenAI Codex, ChatGPT Work, ChatGPT dots | Primary source ↗ |
| Anthropic | 3 profiles | Claude Research, Claude Code, Claude Cowork | Primary source ↗ |
| 7 profiles | Gemini Deep Research, Google Jules, Gemini Spark, Gemini CLI | Primary source ↗ | |
| Google Cloud | 3 profiles | Gemini Enterprise Agent Platform, Gemini Agent, Google AlphaEvolve | Primary source ↗ |
| Microsoft | 7 profiles | Microsoft Copilot Researcher, Microsoft Copilot Studio, Microsoft Copilot Analyst, Microsoft Foundry Agent Service | Primary source ↗ |
| Perplexity | 3 profiles | Perplexity Deep Research, Perplexity Computer, Perplexity Comet Assistant | Primary source ↗ |
| Meta | 1 profiles | Meta Muse | Primary source ↗ |
Which ecosystem fits the task?
Identify the expected action and risk before choosing a provider. A research report, code execution and updating a CRM require different verification.
| Use case | Evidence questions |
|---|---|
| Software development | Compare code execution environment, repository access, review gates and local versus hosted runtime. Examples include OpenAI Codex, Claude Code and Gemini CLI. |
| Research and analysis | Check browsing and citation provenance separately. An agent may produce a sourced report without accessing your enterprise systems. |
| Browser and knowledge work | Examine allowed websites, credentials, external actions, browser isolation, escalation and approval before changes. |
| Enterprise workflows | Identify runtime, connector permissions, user identities, audit export and region/plan availability before purchasing. |
Privacy, approvals and governance
A trustworthy implementation reviews permissions before any external action, human review points, isolation, audit logs, retention, processing regions and customer-data training commitments. These are distinct control families. An official documentation statement is not an independent penetration test. Data residency is not a blanket finding of GDPR compliance; contractual roles, transfers, actual processing purposes and implementation remain relevant.
Evidence method and limitations
AgentenTrust uses 36 published controls with field-level evidence where sources support it. The same company may publish different policies for a workplace agent, coding agent and developer API. Missing data stays Unknown and is never silently converted into a failure. Likewise, the number of indexed agents is a catalogue fact, not a market share or quality score. Read the methodology and compare ecosystems.
Frequently asked questions
Are AI models and AI agents the same?
No. A model produces outputs; an agent adds workflow, tools and a concrete execution surface. Specific rights require product-level evidence.
Does one security control apply to every agent of a provider?
No. AgentenCode records controls for each agent, plan and runtime scope separately.
What does Unknown mean?
It means that the cited sources do not establish the requested field. It is neither No nor Yes.
Are these all agents on the market?
No. The 130 catalogued entries are curated, not a census of the worldwide market.
How should a company evaluate an ecosystem?
Start with the task, access to internal data, approvals, auditability, supported regions and the specific product plan, then review primary sources.
FAQ
Frequently asked questions about AI ecosystems
Are ecosystem agents also in the general directory?
Yes. Ecosystem pages provide another view of the same uniquely identified agent and platform profiles.
How does AgentenCode assess providers?
We show individually documented product facts, evidence coverage and existing AgentenTrust controls, not an unsupported provider-wide rating.
Is an AI model automatically an AI agent?
No. A model is one component; an agent needs a defined execution and tool context.