Integrations

Connect Trefur to the platforms your team already uses. Trefur pulls activity via official admin APIs — no SDK install, no code change in the agents themselves.

How OAuth-based discovery works

  1. In the Trefur dashboard, open Integrations and pick a platform. You see a plain-English explanation of which OAuth scopes will be requested and what each scope grants access to.
  2. Pick the read-only access you want. Trefur asks for the minimum scopes needed to provide the selected visibility.
  3. You're bounced through the platform's OAuth consent screen. We never see your password.
  4. On success, Trefur stores a refresh token in a credential vault and kicks off an instant initial scan. Subsequent polls are incremental.

Supported platforms

PlatformWhat you getSetup
Google WorkspaceAudit logs, Gemini activity, third-party OAuth-grant inventoryWalk-through →
Microsoft 365Audit logs, Copilot prompt countsWalk-through →
OpenAIAdmin API: org + per-user usageWalk-through →
AnthropicAdmin API: workspace + per-key usageWalk-through →
SlackBot + integration activityWalk-through →
AtlassianJira + Confluence agent activityWalk-through →
ZendeskSupport-agent + AI macro activityWalk-through →
HubSpotWorkflow + AI assistant activityWalk-through →
IntercomFin AI + Custom Bot activityWalk-through →
SalesforceEinstein + Agentforce + Flow activityWalk-through →
GCPVertex AI + Gemini API + Cloud Run agent visibilityWalk-through →
AWSBedrock + SageMaker + Lambda visibilityWalk-through →
AzureAzure OpenAI + Cognitive Services + AI Foundry visibilityWalk-through →
SnowflakeCortex AI + Snowpark agent visibilityWalk-through →
DatabricksMLflow + Mosaic AI Agents visibilityWalk-through →

Privacy

Trefur uses read-only OAuth scopes wherever possible. Every credential carries a redaction level that cascades most-restrictive-wins through tenant default → OAuth credential → ingest key → individual request. You cannot accidentally broaden access at a deeper layer.

For prompt-level fidelity (the platform admin APIs don't return prompt + response bodies — that's a platform limitation), install the appropriate SDK or run the collector's LLM proxy in front of outbound model traffic.