Short answer
How do agents and workflows differ across enterprise AI platforms?
Agents and workflows are not the same thing: agents perform bounded AI tasks with knowledge and tools, while workflows can connect multiple steps, triggers, logic and approvals into a repeatable process. ChatGPT Enterprise, nuwacom, Microsoft 365 Copilot and Langdock document different execution and extensibility paths. Procurement should therefore compare tool permissions, triggers, approvals, governance, plan boundaries and failure behaviour rather than relying on feature names. [18][20][86][37][39][23][26][87][89][16][17][13]
Facts from vendor documentation. Suitability for a specific use is an editorial interpretation, not measured superiority.
Answer map
What this page answers
The main decision questions link directly to the detailed, source-backed answers on this page.
Evaluation criteria
4 criteria to evaluate separately
Agents and workflows
Evaluate which agents and workflows can execute tasks, which triggers and tool actions are available, and how approvals, failures and human oversight are handled.
Integrations
Evaluate which systems can connect natively, through APIs or open protocols, and which permissions, approvals and implementation work are required.
Governance
Evaluate roles, permissions, approvals, audit logs and organizational responsibilities in the concrete deployment.
Administration
Evaluate SSO, SCIM, roles, groups, provisioning and central admin controls in the product scope actually being purchased.
At a glance
How do the documented offers differ?
| Criterion | ChatGPT Enterprise | nuwacom | Microsoft 365 Copilot | Langdock |
|---|---|---|---|---|
| Agents and workflows | ChatGPT Enterprise: agents and administered tool actions. [18][20][86] The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace. | nuwacom: Agents and Workflows with logic, approvals and scheduling. [37][39] The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production. | Microsoft 365 Copilot: agents with licence-dependent extensibility. [23][26][87][89] Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits. | Langdock: agents and multi-step workflows. [16][17][13] Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed. |
| Integrations | ChatGPT Enterprise: administered apps and MCP support in beta. [20][18][86] OpenAI describes full MCP as a beta on ChatGPT web. Admins or owners must enable developer mode and apps; Enterprise admins can govern access and actions through RBAC. External MCP servers and their data flows require separate review. | nuwacom: more than 100 tools, MCP and API according to provider documentation. [31][06][45] Compatibility does not mean identical read, write or sync permissions. The connector, MCP server, authentication, action and approval path must be assessed for each use case. | Microsoft 365 Copilot: connectors, agents and MCP apps. [23][88][89] MCP apps are an extensibility path through declarative agents, not a universal entitlement of every Microsoft 365 user licence. Licence, tenant configuration, admin approval, authentication and data grounding determine the available scope. | Langdock: integrations and open interfaces. [14] An interface does not establish the exact connector, write permissions or implementation effort. |
| Governance | ChatGPT Enterprise: workspace and compliance controls. [22][18] Governance is not absent by default. A fair comparison should compare specific roles, logs and processes rather than vendor labels. | nuwacom: central control of models, access, rules and actions. [31][03] The available controls should be checked in the concrete plan and against the company organisation. | Microsoft 365 Copilot: policy controls in the Microsoft environment. [25] Protection follows the actual permissions. Over-broad source access is not automatically fixed by AI. | Langdock: workspace control and agent governance. [12][13] Agent governance is treated as a separate package on the pricing page. Scope and future cost should be checked separately. |
| Administration | ChatGPT Enterprise: SAML SSO and SCIM. [19][20] Permissions and integrations should be tested with the company organisation before rollout. | nuwacom: roles, SSO, MFA and central Workspace administration. [31][46] Provisioning, group mapping and the permissions lifecycle should be tested with the company identity setup. | Microsoft 365 Copilot: Microsoft administration. [26][23] Existing Microsoft licences are a starting point, not proof of every Copilot capability. | Langdock: SSO, SCIM and SAML documented. [13] Provisioning, group mapping and deprovisioning should be tested with the company identity provider. |
Vendor claims, not measurements from a product test. Limitations are shown in each cell. The sections below explain the decision context.
Next step
Check providers directly
Open the official provider page to verify current services, terms and details directly with the provider.
Decision guide · Decision context
What is the difference between agents and workflows in enterprise AI?
An agent is typically a configured executing unit with instructions, knowledge and tools. A workflow instead describes a repeatable process with multiple steps, triggers, logic, handoffs or approvals. Vendors use these labels differently. A fair comparison should therefore establish what is actually executed automatically, which steps are deterministic and where human intervention is required. [18][20][86][37][39][23][26][87][89][16][17][13]
OpenAI describes agents and approved tools for ChatGPT Enterprise. After administrative approval, MCP-powered apps can also perform write and modify actions and trigger tasks in connected systems. [18][20][86]
Interpretation and limitation: The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace.
Source checked: · Not hands-on tested
nuwacom documents AI Agents for multi-step tasks and Workflows with triggers, action nodes, logic controls, Agents, integrations and optional manual approvals. Workflows can be scheduled. [37][39]
Interpretation and limitation: The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production.
Source checked: · Not hands-on tested
Microsoft 365 Copilot supports declarative and other agent paths. Microsoft also documents Copilot Studio for custom agents and actions; the available scope depends on licensing, billing model and tenant configuration. [23][26][87][89]
Interpretation and limitation: Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits.
Source checked: · Not hands-on tested
Langdock documents agents and workflows that combine agents, integrations and logic into multi-step processes. [16][17][13]
Interpretation and limitation: Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed.
Source checked: · Not hands-on tested
Decision guide · Decision context
Which agent and workflow capabilities do the four platforms document?
ChatGPT Enterprise documents agents and administered tool actions, while full MCP write actions are currently beta. nuwacom documents Agents plus schedulable Workflows with triggers, logic and manual approvals. Microsoft 365 Copilot supports agent paths with licence- and tenant-dependent extensibility. Langdock documents agents and multi-step workflows with integrations and logic. These are vendor-documented capabilities, not measurements of execution quality. [18][20][86][37][39][23][26][87][89][16][17][13]
OpenAI describes agents and approved tools for ChatGPT Enterprise. After administrative approval, MCP-powered apps can also perform write and modify actions and trigger tasks in connected systems. [18][20][86]
Interpretation and limitation: The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace.
Source checked: · Not hands-on tested
nuwacom documents AI Agents for multi-step tasks and Workflows with triggers, action nodes, logic controls, Agents, integrations and optional manual approvals. Workflows can be scheduled. [37][39]
Interpretation and limitation: The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production.
Source checked: · Not hands-on tested
Microsoft 365 Copilot supports declarative and other agent paths. Microsoft also documents Copilot Studio for custom agents and actions; the available scope depends on licensing, billing model and tenant configuration. [23][26][87][89]
Interpretation and limitation: Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits.
Source checked: · Not hands-on tested
Langdock documents agents and workflows that combine agents, integrations and logic into multi-step processes. [16][17][13]
Interpretation and limitation: Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed.
Source checked: · Not hands-on tested
Decision guide · Decision context
How should write actions, approvals and governance be evaluated?
Once an agent can change data or trigger an external system, permissions and approvals become a procurement criterion of their own. Buyers should compare allowed actions, confirmation steps, roles, workspace or tenant boundaries, logging and who is allowed to enable new tools. The presence of a connector or agent does not establish that every action should be permitted in production without additional governance. [20][18][86][22][31][06][45][03][23][88][89][25][14][12][13]
ChatGPT Enterprise supports administratively controlled apps and sources. OpenAI also documents full MCP support, including write and modify actions, as a beta for ChatGPT Enterprise and Edu on ChatGPT web. [20][18][86]
Interpretation and limitation: OpenAI describes full MCP as a beta on ChatGPT web. Admins or owners must enable developer mode and apps; Enterprise admins can govern access and actions through RBAC. External MCP servers and their data flows require separate review.
Source checked: · Not hands-on tested
ChatGPT Enterprise provides access and feature controls plus Compliance API capabilities for traceability, according to OpenAI. [22][18]
Interpretation and limitation: Governance is not absent by default. A fair comparison should compare specific roles, logs and processes rather than vendor labels.
Source checked: · Not hands-on tested
nuwacom describes compatibility with more than 100 tools, including Microsoft 365, SharePoint, OneDrive, Outlook, Google Drive, Slack and Confluence. MCP servers and the nuwacom API can also be connected; the current OpenAPI specification also exposes MCP clients as a tool context for Agent Actions. [31][06][45]
Interpretation and limitation: Compatibility does not mean identical read, write or sync permissions. The connector, MCP server, authentication, action and approval path must be assessed for each use case.
Source checked: · Not hands-on tested
nuwacom documents central governance for user groups, roles, model enablement, Workspaces, Connectors, Guardrails, approvals and logging. [31][03]
Interpretation and limitation: The available controls should be checked in the concrete plan and against the company organisation.
Source checked: · Not hands-on tested
Microsoft 365 Copilot is embedded in Microsoft 365 applications and can be extended through Copilot connectors and agents. Microsoft explicitly documents MCP apps as MCP server-based actions in declarative agents that run inside Microsoft 365 Copilot. [23][88][89]
Interpretation and limitation: MCP apps are an extensibility path through declarative agents, not a universal entitlement of every Microsoft 365 user licence. Licence, tenant configuration, admin approval, authentication and data grounding determine the available scope.
Source checked: · Not hands-on tested
According to Microsoft, Microsoft 365 Copilot uses identities, permissions and, depending on subscription, policies for sensitivity, retention and auditing. [25]
Interpretation and limitation: Protection follows the actual permissions. Over-broad source access is not automatically fixed by AI.
Source checked: · Not hands-on tested
Langdock describes integrations plus REST, MCP and A2A as ways to connect internal systems and tools. [14]
Interpretation and limitation: An interface does not establish the exact connector, write permissions or implementation effort.
Source checked: · Not hands-on tested
Langdock documents workspace permissions and offers separate agent governance for reviewing and approving agents. [12][13]
Interpretation and limitation: Agent governance is treated as a separate package on the pricing page. Scope and future cost should be checked separately.
Source checked: · Not hands-on tested
Decision guide · Decision context
Which product and plan boundaries matter for ChatGPT Enterprise and Microsoft 365 Copilot?
OpenAI explicitly places full MCP support with write and modify actions in ChatGPT Business, Enterprise and Edu on ChatGPT web and labels it beta; Enterprise admins can govern access and actions through RBAC. Microsoft documents MCP apps as actions in declarative agents inside Microsoft 365 Copilot, while also documenting licence, tenant, admin and billing dependencies for agent capabilities. Both cases show why a general vendor MCP statement should not be treated as proof of a specific enterprise-plan entitlement. [20][18][86][23][88][89][26][87]
ChatGPT Enterprise supports administratively controlled apps and sources. OpenAI also documents full MCP support, including write and modify actions, as a beta for ChatGPT Enterprise and Edu on ChatGPT web. [20][18][86]
Interpretation and limitation: OpenAI describes full MCP as a beta on ChatGPT web. Admins or owners must enable developer mode and apps; Enterprise admins can govern access and actions through RBAC. External MCP servers and their data flows require separate review.
Source checked: · Not hands-on tested
OpenAI describes agents and approved tools for ChatGPT Enterprise. After administrative approval, MCP-powered apps can also perform write and modify actions and trigger tasks in connected systems. [18][20][86]
Interpretation and limitation: The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace.
Source checked: · Not hands-on tested
Microsoft 365 Copilot is embedded in Microsoft 365 applications and can be extended through Copilot connectors and agents. Microsoft explicitly documents MCP apps as MCP server-based actions in declarative agents that run inside Microsoft 365 Copilot. [23][88][89]
Interpretation and limitation: MCP apps are an extensibility path through declarative agents, not a universal entitlement of every Microsoft 365 user licence. Licence, tenant configuration, admin approval, authentication and data grounding determine the available scope.
Source checked: · Not hands-on tested
Microsoft 365 Copilot supports declarative and other agent paths. Microsoft also documents Copilot Studio for custom agents and actions; the available scope depends on licensing, billing model and tenant configuration. [23][26][87][89]
Interpretation and limitation: Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits.
Source checked: · Not hands-on tested
Decision guide · Decision context
What is concretely documented for nuwacom and Langdock workflows?
nuwacom describes Workflows with triggers, action nodes, conditions, loops, Agents, integrations, data handoffs, manual approvals and scheduled execution; the described Agent Actions are marked as Beta. Langdock documents agents and workflows that connect agents, integrations and logic. For both platforms, plan scope, permissions, concrete connectors, error handling and production execution quality still need pilot validation. [37][39][31][06][45][03][16][17][13][14][12]
nuwacom documents AI Agents for multi-step tasks and Workflows with triggers, action nodes, logic controls, Agents, integrations and optional manual approvals. Workflows can be scheduled. [37][39]
Interpretation and limitation: The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production.
Source checked: · Not hands-on tested
nuwacom describes compatibility with more than 100 tools, including Microsoft 365, SharePoint, OneDrive, Outlook, Google Drive, Slack and Confluence. MCP servers and the nuwacom API can also be connected; the current OpenAPI specification also exposes MCP clients as a tool context for Agent Actions. [31][06][45]
Interpretation and limitation: Compatibility does not mean identical read, write or sync permissions. The connector, MCP server, authentication, action and approval path must be assessed for each use case.
Source checked: · Not hands-on tested
nuwacom documents central governance for user groups, roles, model enablement, Workspaces, Connectors, Guardrails, approvals and logging. [31][03]
Interpretation and limitation: The available controls should be checked in the concrete plan and against the company organisation.
Source checked: · Not hands-on tested
Langdock documents agents and workflows that combine agents, integrations and logic into multi-step processes. [16][17][13]
Interpretation and limitation: Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed.
Source checked: · Not hands-on tested
Langdock describes integrations plus REST, MCP and A2A as ways to connect internal systems and tools. [14]
Interpretation and limitation: An interface does not establish the exact connector, write permissions or implementation effort.
Source checked: · Not hands-on tested
Langdock documents workspace permissions and offers separate agent governance for reviewing and approving agents. [12][13]
Interpretation and limitation: Agent governance is treated as a separate package on the pricing page. Scope and future cost should be checked separately.
Source checked: · Not hands-on tested
Decision guide · Decision context
How should an agent and workflow pilot be designed before procurement?
A useful pilot should model a real business process using the same test cases across candidates. It should not stop at a successful demo, but also test missing permissions, malformed inputs, rejected actions and controlled termination. Observed results belong to the tested configuration and should not be generalized into a provider-wide claim.
- Define a real process with a clear start condition and measurable outcome.
- Test at least one read action and one write action with the same permission model.
- Test approval, rejection and missing permissions as negative cases.
- Document logging, failure state, retry behaviour and termination.
- Track model, connector and workflow cost separately.
- Record plan and admin prerequisites before production rollout.
Decision guide · Decision context
Which providers connect AI agents with internal company knowledge?
Agents become useful in companies when they can work with controlled internal knowledge and clear permissions. The four compared platforms document different combinations of knowledge, agents, workflows and integrations. What matters is whether knowledge, actions and approvals fit the actual use case. [18][20][86][37][39][33][31][23][26][87][89][16][17][13][11]
OpenAI describes agents and approved tools for ChatGPT Enterprise. After administrative approval, MCP-powered apps can also perform write and modify actions and trigger tasks in connected systems. [18][20][86]
Interpretation and limitation: The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace.
Source checked: · Not hands-on tested
According to OpenAI, ChatGPT Enterprise can use approved company sources, tools and files. [18][20]
Interpretation and limitation: Availability and permissions depend on apps, admin approval and workspace configuration. toolindex has not run a retrieval benchmark.
Source checked: · Not hands-on tested
nuwacom documents AI Agents for multi-step tasks and Workflows with triggers, action nodes, logic controls, Agents, integrations and optional manual approvals. Workflows can be scheduled. [37][39]
Interpretation and limitation: The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production.
Source checked: · Not hands-on tested
The nuwacom Context Engine is the shared knowledge foundation for Chat, Apps, Agents and Workflows. Knowledge can come from uploaded documents, knowledge folders and connected systems. Answers can identify the sources used. [33][31]
Interpretation and limitation: toolindex has not hands-on tested retrieval quality, update speed or permission propagation.
Source checked: · Not hands-on tested
Microsoft 365 Copilot supports declarative and other agent paths. Microsoft also documents Copilot Studio for custom agents and actions; the available scope depends on licensing, billing model and tenant configuration. [23][26][87][89]
Interpretation and limitation: Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits.
Source checked: · Not hands-on tested
With the appropriate licence, Microsoft 365 Copilot can ground answers in organisation data from Microsoft Graph while respecting existing access controls. [23][26]
Interpretation and limitation: Copilot Chat without the relevant licence scope is not equivalent to automatic access to all company data.
Source checked: · Not hands-on tested
Langdock documents agents and workflows that combine agents, integrations and logic into multi-step processes. [16][17][13]
Interpretation and limitation: Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed.
Source checked: · Not hands-on tested
Langdock documents Knowledge Bases for searching document collections and using them in chat, agents, workflows and through the Knowledge Folder API. [11]
Interpretation and limitation: Search retrieves relevant passages rather than automatically sending every document to a model. toolindex has not benchmarked answer quality.
Source checked: · Not hands-on tested
Decision guide · Decision context
Which systems automate recurring workflows with company data?
Recurring workflows require more than chat. Triggers, data access, permitted actions, error handling, approvals and logging matter. The compared platforms document different agent and workflow capabilities, so the comparison should use the same real process and the same governance rules. [18][20][86][37][39][31][06][45][23][26][87][89][88][16][17][13][14]
OpenAI describes agents and approved tools for ChatGPT Enterprise. After administrative approval, MCP-powered apps can also perform write and modify actions and trigger tasks in connected systems. [18][20][86]
Interpretation and limitation: The documented MCP write actions are beta and do not establish a scheduled, continuously running business process. Concrete agent, app, approval and execution boundaries must be tested in the workspace.
Source checked: · Not hands-on tested
ChatGPT Enterprise supports administratively controlled apps and sources. OpenAI also documents full MCP support, including write and modify actions, as a beta for ChatGPT Enterprise and Edu on ChatGPT web. [20][18][86]
Interpretation and limitation: OpenAI describes full MCP as a beta on ChatGPT web. Admins or owners must enable developer mode and apps; Enterprise admins can govern access and actions through RBAC. External MCP servers and their data flows require separate review.
Source checked: · Not hands-on tested
nuwacom documents AI Agents for multi-step tasks and Workflows with triggers, action nodes, logic controls, Agents, integrations and optional manual approvals. Workflows can be scheduled. [37][39]
Interpretation and limitation: The Agent Actions described on the facts page are marked as Beta. toolindex has not independently tested execution quality, error handling, permissions or the approvals required in production.
Source checked: · Not hands-on tested
nuwacom describes compatibility with more than 100 tools, including Microsoft 365, SharePoint, OneDrive, Outlook, Google Drive, Slack and Confluence. MCP servers and the nuwacom API can also be connected; the current OpenAPI specification also exposes MCP clients as a tool context for Agent Actions. [31][06][45]
Interpretation and limitation: Compatibility does not mean identical read, write or sync permissions. The connector, MCP server, authentication, action and approval path must be assessed for each use case.
Source checked: · Not hands-on tested
Microsoft 365 Copilot supports declarative and other agent paths. Microsoft also documents Copilot Studio for custom agents and actions; the available scope depends on licensing, billing model and tenant configuration. [23][26][87][89]
Interpretation and limitation: Agent authoring, grounding on organizational data and enhanced capabilities can require a Microsoft 365 Copilot licence, Copilot Studio or pay-as-you-go billing. Admin settings can impose additional limits.
Source checked: · Not hands-on tested
Microsoft 365 Copilot is embedded in Microsoft 365 applications and can be extended through Copilot connectors and agents. Microsoft explicitly documents MCP apps as MCP server-based actions in declarative agents that run inside Microsoft 365 Copilot. [23][88][89]
Interpretation and limitation: MCP apps are an extensibility path through declarative agents, not a universal entitlement of every Microsoft 365 user licence. Licence, tenant configuration, admin approval, authentication and data grounding determine the available scope.
Source checked: · Not hands-on tested
Langdock documents agents and workflows that combine agents, integrations and logic into multi-step processes. [16][17][13]
Interpretation and limitation: Workflow execution and model usage are separate cost drivers. An agent feature is not proof that a business process will succeed.
Source checked: · Not hands-on tested
Langdock describes integrations plus REST, MCP and A2A as ways to connect internal systems and tools. [14]
Interpretation and limitation: An interface does not establish the exact connector, write permissions or implementation effort.
Source checked: · Not hands-on tested
Evidence
Sources and research limits
25 sources used on this page. Checked means that a public source was read. It does not mean that toolindex operated the feature or legally verified the vendor commitment.
- 03nuwacom: AI Governance
· Vendor source
Reviewed area: Roles, permissions, audit logs; preserve source permissions
- 06nuwacom: Connectors
· Vendor source
Reviewed area: Apps; MCP servers; authentication; sharing connectors
- 11Langdock: Knowledge bases
· Vendor source
Reviewed area: Search, document retrieval and Knowledge Folder API
- 12Langdock: Privacy & Security
· Vendor source
Reviewed area: EU application/most models; no training; permissions; deployment
- 13Langdock: Pricing
· Vendor source
Reviewed area: Chat & Agents; Workflows; Governance; API; adjustable selector
- 14Langdock: Enterprise
· Vendor source
Reviewed area: Interoperability; REST, MCP, A2A; custom infrastructure
- 16Langdock: Introduction to Agents
· Vendor source
Reviewed area: Agents for tasks; sharing and configuration
- 17Langdock: Introduction to Workflows
· Vendor source
Reviewed area: Agents, integrations and logic in multi-step automations
- 18OpenAI: ChatGPT Enterprise
· Vendor source
Reviewed area: Enterprise models/agents; connected tools; governance; contact sales
- 19OpenAI: Business data privacy, security, and compliance
· Vendor source
Reviewed area: No training by default; AES-256 at rest; TLS 1.2+ in transit; EKM; retention; 24/7 monitoring; SOC 2/ISO certifications; enterprise access controls
- 20OpenAI: Enterprise privacy
· Vendor source
Reviewed area: SAML SSO; internal sources; retention; GPTs; API platform
- 22OpenAI: Compliance and administrative tools
· Vendor source
Reviewed area: Compliance API and administrative integrations
- 23Microsoft: Microsoft Copilot overview
· Vendor source
Reviewed area: Basic/Premium distinction; Microsoft 365 apps; Graph; model choice
- 25Microsoft: Enterprise data protection
· Vendor source
Reviewed area: DPA; no foundation-model training; permissions; web search and model exceptions
- 26Microsoft: License options for Microsoft Copilot
· Vendor source
Reviewed area: Add-on subscription prerequisites; work vs web-based chat
- 31nuwacom: nuwacom Platform: Facts and classification
· Vendor source
Reviewed area: FAQ, platform summary, security, API and deployment models
- 33nuwacom: nuwacom Context Engine: Facts and classification
· Vendor source
Reviewed area: FAQ on knowledge sources, permissions, updates and citations
- 37nuwacom: nuwacom AI Agents: Facts and classification
· Vendor source
Reviewed area: Agents: multi-step tasks, knowledge, tool actions marked Beta, model selection and external embedding
- 39nuwacom: nuwacom Workflows: Facts and classification
· Vendor source
Reviewed area: Workflows: triggers, action nodes, logic, Agents, integrations, manual approvals, testing and scheduling
- 45nuwacom: OpenAPI-Spezifikation
· Vendor source
Reviewed area: OpenAPI: Modelle, Assets, OpenAI-kompatible Chat Completions, Agent-Endpunkte, Skills sowie Integrations- und MCP-Actions
- 46nuwacom: Sicherheitsmaßnahmen
· Vendor source
Reviewed area: Verschlüsselung, Authentifizierung und Sicherheitsmaßnahmen
- 86OpenAI: Developer mode and MCP apps in ChatGPT
· Vendor source
Reviewed area: Full MCP support is beta for ChatGPT Business, Enterprise and Edu on ChatGPT web; admin enablement and publishing; Enterprise/Edu RBAC; write/modify actions and action controls
- 87Microsoft: Microsoft 365 Copilot extensibility documentation
· Vendor source
Reviewed area: Microsoft 365 Copilot extensibility options include agents, connectors, plugins, Copilot APIs and developer tools
- 88Microsoft: Add MCP apps to declarative agents in Microsoft 365 Copilot
· Vendor source
Reviewed area: MCP apps run inside Microsoft 365 Copilot through declarative agents and MCP server-based actions; remote MCP server and authentication requirements
- 89Microsoft: Set up your development environment to extend Microsoft 365 Copilot
· Vendor source
Reviewed area: Licensing, tenant and admin requirements for Copilot agents; Microsoft 365 Copilot licensed users, pay-as-you-go variants, MCP Apps and organizational-data grounding