AgentProof

workspace · environments · setup

Connect an environment

Pick a connector. Microsoft is the first real connector; the rest are clearly labelled planned.

Connect Microsoft →Azure AI FoundryAdd a non-Microsoft agent manuallyOpen the demo

Connector state · classified by R52 microsoft_connector_state_v2.

Workspace · Environments · Setup wizard

Connector-first environment setup

Pick the platform that hosts your agents. AgentProof handles the connection inside the product UI. Manual fallback is available where the connector path has not yet shipped — but it is a fallback, not the default.

Connector-first setup

Pick a connector

Connectors are listed by readiness. Real connectors first, then simulated-discovery previews, then planned. Manual fallback is the last resort.

Available: 1 · Simulated: 2 · Planned: 7 · Manual fallback: 1

  • Microsoft Power Platform / Copilot Studio

    discovery_only_simulated

    setup: oauth_connector · auth: microsoft_entra_oauth · fallback: simulated_discovery

    Discovery objects: environments, copilot_studio_agents, topics, actions

    Next build action: Wire the existing /api/connectors/microsoft/* routes into the connector-first setup wizard so customers no longer paste an environment name.

  • Azure AI Foundry

    discovery_only_simulated

    setup: oauth_connector · auth: microsoft_entra_oauth · fallback: simulated_discovery

    Discovery objects: projects, workspaces, deployments, agents

    Next build action: Implement the real Microsoft.MachineLearningServices/workspaces read endpoint + promote to real_connector_shipped.

  • OpenAI

    connector_planned

    setup: manual_paste · auth: openai_api_key_per_workspace · fallback: manual_form

    Discovery objects: models, assistants, actions

    Next build action: Ship the OpenAI Assistants/read connector + workspace-scoped API key intake under RLS.

  • Anthropic

    connector_planned

    setup: manual_paste · auth: anthropic_api_key_per_workspace · fallback: manual_form

    Discovery objects: models, prompts

    Next build action: Ship the Anthropic model/read connector + workspace-scoped API key intake under RLS.

  • Google Vertex AI

    connector_planned

    setup: oauth_connector · auth: google_oauth · fallback: manual_form

    Discovery objects: models, endpoints, agents

    Next build action: Ship the Google OAuth + Vertex AI Endpoints/list connector.

  • AWS Bedrock

    connector_planned

    setup: oauth_connector · auth: aws_iam_role · fallback: manual_form

    Discovery objects: foundation_models, agents, knowledge_bases

    Next build action: Ship the AWS Bedrock ListAgents connector backed by IAM role assumption.

  • ServiceNow

    connector_planned

    setup: oauth_connector · auth: servicenow_oauth · fallback: manual_form

    Discovery objects: virtual_agents, now_assist_skills, flows

    Next build action: Ship the ServiceNow OAuth + Virtual Agent metadata connector.

  • Salesforce (Einstein / Agentforce)

    connector_planned

    setup: oauth_connector · auth: salesforce_oauth · fallback: manual_form

    Discovery objects: einstein_copilots, agentforce_agents, flows

    Next build action: Ship the Salesforce OAuth + Agentforce metadata connector.

  • Self-hosted frameworks (LangChain / CrewAI / AutoGen)

    connector_planned

    setup: self_hosted_smoke · auth: none_self_hosted · fallback: manual_form

    Discovery objects: agent_graph_export, tool_inventory

    Next build action: Ship a self-hosted smoke that the customer can run to export an agent graph + tool inventory for AgentProof to consume.

  • Manual (fallback)fallback

    real_connector_shipped

    setup: manual_form · auth: none_manual · fallback: none

    Discovery objects: (none — manual)

    Next build action: Keep manual fallback live but visibly secondary on the setup chooser.

Connector-first environment setup

10 connector paths · 0 live · 1 blocked (real connector ready, customer tenant required) · 1 simulated · 6 planned · 2 manual fallback.

Discovery is the primary path. Manual fallback is for cases the connector doesn't cover yet.

  • Microsoft Power Platform / Copilot Studio

    real connector blocked
    Can discover:
    Environments, Copilot Studio agents, agent configuration metadata.
    Permissions:
    Read-only access to Power Platform admin API + Copilot Studio agents. Requires Microsoft Entra ID app + tenant admin consent.
    AgentProof reads:
    Agent metadata, environment list, agent configuration (read-only).
    AgentProof does NOT read:
    Business records, customer conversations, secrets.
    Configure Microsoft tenant →

  • Azure AI Foundry

    simulated clearly labelled
    Can discover:
    AI projects, agents, model registry (when live).
    Permissions:
    Read-only Azure AI Foundry tenant permissions.
    AgentProof reads:
    Project metadata, agent definitions (read-only).
    AgentProof does NOT read:
    Project secrets, raw model weights, customer data.
    Open Foundry setup →

  • OpenAI Agents

    planned not started
    Can discover:
    OpenAI Assistant API agents, threads, tools (when live).
    Permissions:
    OpenAI organisation API key with read-only scope.
    AgentProof reads:
    Assistant metadata, tool definitions, model selection.
    AgentProof does NOT read:
    Thread content, customer conversations, secrets.
    Mark as planned →

  • Anthropic Agents

    planned not started
    Can discover:
    Claude agent configurations, tool definitions (when live).
    Permissions:
    Anthropic organisation API key with read-only scope.
    AgentProof reads:
    Agent configuration, system prompts, tool definitions.
    AgentProof does NOT read:
    Customer conversations, prompts, secrets.
    Mark as planned →

  • Google Vertex AI Agent Builder

    planned not started
    Can discover:
    Vertex AI agents, data stores, conversation flows (when live).
    Permissions:
    Google Cloud read-only IAM role.
    AgentProof reads:
    Agent metadata, data store schema (read-only).
    AgentProof does NOT read:
    Customer documents, conversation content.
    Mark as planned →

  • AWS Bedrock Agents

    planned not started
    Can discover:
    Bedrock agents, knowledge bases, action groups (when live).
    Permissions:
    AWS IAM read-only role for Bedrock.
    AgentProof reads:
    Agent definition, knowledge base names, model selection.
    AgentProof does NOT read:
    Knowledge base content, customer data.
    Mark as planned →

  • ServiceNow AI Agents

    planned not started
    Can discover:
    ServiceNow Now Assist + AI agents (when live).
    Permissions:
    ServiceNow read-only admin role.
    AgentProof reads:
    Agent configuration, integration points.
    AgentProof does NOT read:
    Customer tickets, sensitive ITSM data.
    Mark as planned →

  • Salesforce Agentforce

    planned not started
    Can discover:
    Agentforce agents, flows, action sets (when live).
    Permissions:
    Salesforce read-only API user.
    AgentProof reads:
    Agent metadata, flow definitions.
    AgentProof does NOT read:
    CRM records, customer PII.
    Mark as planned →

  • Open-source agent frameworks (LangChain / CrewAI / AutoGen)

    manual fallback
    Can discover:
    Manually-pasted agent configuration (no live API).
    Permissions:
    None — paste your agent config into the form.
    AgentProof reads:
    Whatever you paste — labelled manual.
    AgentProof does NOT read:
    Anything you don't paste.
    Use manual fallback →

  • Custom / manual fallback

    manual fallback
    Can discover:
    Whatever you manually describe to AgentProof.
    Permissions:
    None.
    AgentProof reads:
    Only what you provide.
    AgentProof does NOT read:
    Anything else.
    Open manual setup →

How this works

  • Read-only by default. No business data extracted.
  • Least-privilege permissions only.
  • No source-code access required.
  • No broad tenant scanning. You choose the scope.
  • No secret values ever displayed.