LLM Connectors — overview

Claresia is platform-portable. The same Skill IR ships into all four enterprise LLM platforms via the Distribution Plane. You pick whichever your organization already standardized on; Claresia publishes the catalog and ingests the audit log.

Supported platforms

Platform	Roadmap	Distribution mechanism	Telemetry source
Anthropic Claude Enterprise	cc-063	Anthropic Admin API → publishes as Claude Skills	Anthropic Audit API
Microsoft Copilot M365	cc-065	Power Platform Admin API → Copilot Studio agents + Power Platform actions	Microsoft Graph audit log
OpenAI ChatGPT Enterprise	cc-070	OpenAI Compliance API → Custom GPTs + tools (or manual upload)	OpenAI Compliance API
Google Gemini for Workspace	cc-070+	Google Workspace Admin API → Gemini Gems	Workspace Admin Audit API

You can connect multiple platforms simultaneously — common pattern is “Claude for engineering, Copilot for everyone else.”

Common requirements (all platforms)

• An enterprise tier of the LLM (not the consumer / team tier — Claresia needs the admin/compliance/audit surfaces)
• A service account or admin API key with scopes documented per platform
• Zero-retention enabled in the platform (Claresia only routes through zero-retention modes; we will refuse to publish to a tenant where retention is on)

What the Distribution Plane does

For each connected platform, Claresia:

1. Reads your tenant’s Skill IR catalog (versioned, RBAC-filtered per archetype)
2. Transpiles each skill into the platform’s native shape (Claude Skill JSON, Copilot Studio agent YAML, OpenAI Custom GPT bundle, Gemini Gem)
3. Publishes via the platform’s admin API under the claresia-{tenant_slug}-* namespace
4. Sets RBAC to match your group → archetype mapping (where the platform supports per-user / per-group skill scoping)
5. Re-publishes on change — if you toggle a skill in the Onboarding Portal, the publish lag is <60s p99

What the Telemetry Pipeline does

For each connected platform:

1. Pulls the audit log on a 60s polling cadence (or webhook where the platform supports it)
2. Filters for events tagged claresia-*
3. Normalizes into fn_telemetry_event shape (skill_id, ts, success, latency_ms, tokens_in, tokens_out, cost_usd_estimate, …)
4. Writes to the Hub telemetry_event table
5. Surfaces in Command Center within 5 min p95 of the LLM invocation

Per-platform deep dives

• Claude Enterprise
• Microsoft Copilot M365
• OpenAI ChatGPT Enterprise
• Google Gemini for Workspace

When the platform doesn’t have an admin API

Some platforms (notably ChatGPT Enterprise on certain SKUs) do not yet expose a fully programmatic publish API. Claresia falls back to:

• Manual-upload UX in Command Center: you download a per-skill ZIP bundle
  • a CSV manifest, upload it via the platform’s admin UI in one batch
• Per-skill instruction copy-paste for the Custom GPT builder
• CSV import for ChatGPT Enterprise’s bulk Custom GPT manager (if enabled)

This fallback is documented per-platform.

Skills lifecycle (across platforms)

graph LR
    A[Author Skill IR] --> B[Validate against schema]
    B --> C{Distribution Plane}
    C -->|cc-063| D[Claude Enterprise]
    C -->|cc-065| E[Microsoft Copilot]
    C -->|cc-070| F[ChatGPT Enterprise]
    C -->|cc-070+| G[Google Gemini]
    D --> H[End-user invokes @claresia.X]
    E --> H
    F --> H
    G --> H
    H --> I[Output + telemetry → Hub]
    I --> J[Command Center surfaces]

What the end user sees

Regardless of which LLM your org uses, the user types @claresia.<skill-name> in their LLM, gets a parameter prompt, gets the response wrapped in a Claresia Adaptive Card with the Hub deep-link footer. The experience is intentionally near-identical across all four platforms — that’s the point of the IR.

See End-user guides for archetype-specific walkthroughs.