Microsoft Copilot Studio
Custom Connector imported from OpenAPI, Entra ID OAuth so the signed-in user's identity flows through to SAP/IAM. ~15 min setup.
Read setup guideMicrosoft Copilot, SAP Joule, Anthropic Claude, ChatGPT, LangChain, LlamaIndex — they all hit the same nine-stage AegisAI gateway. Same auth. Same trust ledger. Same masking firewall. Same audit chain. The only thing that changes per platform is how the agent platform discovers the API and authenticates the user.
Each connector ships with a ready-to-import OpenAPI/manifest file plus a step-by-step README. Setup time is in the 5–30 minute range per platform.
Custom Connector imported from OpenAPI, Entra ID OAuth so the signed-in user's identity flows through to SAP/IAM. ~15 min setup.
Read setup guideCustom Action via a BTP destination. SAML bearer assertion propagates the SAP user. Useful when you want one audit chain that covers Joule and everything else. ~30 min setup.
Read setup guideTool-use spec for the Anthropic SDK. Works with Claude Desktop, Claude Code, Bedrock, and Vertex. ~5 min setup.
Read setup guideOpenAPI Action for Custom GPTs and ChatGPT Enterprise workspace connectors. SSO-aware. ~10 min setup.
Read setup guideBaseTool subclass — drop into any LangChain agent (ReAct, OpenAI Functions, LangGraph). ~80 lines, no install. ~5 min setup.
FunctionTool factory for any LlamaIndex agent (ReActAgent, OpenAIAgent, Workflows). ~5 min setup.
Every connector here forwards the end user's bearer JWT to AegisAI. The agent platform never gets to mint its own service token and use it on behalf of the user. AegisAI's audit row records the human's sub — not "Copilot," not "ChatGPT," not "the bot."
Whether the request came from Joule, Claude, or a homegrown LangChain agent, it lands in the same Postgres-backed HMAC chain. Your SOC analyst re-walks one log, not seven.
The same priority-weighted expression set evaluates every request. A field tag that's restricted for one agent is restricted for all of them. There is no "Copilot exception."
Per-field classification × user clearance × auth-object gate. Fields the user can't see come back blanked, regardless of which agent asked.
Speaks OpenAPI? Most agent platforms (Power Automate, Zapier, n8n, custom Action frameworks) import an OpenAPI 3.0 spec. Use ours: samples/connectors/openapi/aegisai.yaml.
Speaks Python or plain HTTP? Look at the LangChain tool — it's about 60 lines. The same shape works in any language: POST /query with {token, intent, context}, parse the response, surface the trace_id to your audit team.
If you build a connector for a platform we haven't covered, we'll merge it back into the repo and add it to this catalogue.