Inference Enclave
Private inference runtime for Luca — hosted VPC or on-prem appliance, with tenant data kept out of training.
What is Inference Enclave?
Inference Enclave (product name Enclave in the UI) is TripleBooks’s private inference runtime for Luca. It runs model inference for chat, classification, and tool-backed workflows inside an environment your organization controls — a dedicated hosted deployment or an on-premises appliance — so prompts and tenant context stay on infrastructure aligned with your security and residency requirements.
Enclave is how Luca answers and acts when you need stronger isolation than the default shared inference path used on TripleBooks Cloud.
Hosted and appliance
| Delivery | Where it runs | Typical fit |
|---|---|---|
| Hosted Enclave | Dedicated capacity in a VPC or isolated cloud tenancy managed with TripleBooks | Enterprise workspaces that want private inference without operating hardware |
| Appliance | On-premises or customer-managed rack / VM images | Regulated or air-gapped environments that keep inference inside their network |
Both options serve the same product contract: Luca tools and policies still apply; only the inference plane moves into the private runtime.
Contact enterprise sales for sizing, residency, and licensing.
Isolation and data use
- Tenant-scoped context — Luca still resolves identity, entity, and role before tools run; Enclave receives only what that session is allowed to see.
- No training on your tenant data — prompts, documents, and ledger context processed through Enclave are not used to train foundation models.
- Encryption and audit continuity — conversation storage, field-level encryption, and audit trails remain the same product surfaces described under Encryption and Architecture.
How Luca uses Enclave
When Enclave is enabled for a workspace, Luca routes model calls through the private runtime for in-product chat, agents, and classification helpers. Tool execution (reads, proposals, posting) continues through TripleBooks APIs with the same RBAC and credit or self-hosted metering rules as today.
Operators configure Enclave through enterprise onboarding (endpoint, credentials, and health checks). Day-to-day users still open Luca — the runtime change is infrastructure, not a separate chat product.
See Luca AI for agents, guardrails, and where Luca lives in the app.
Compared with default shared inference
On TripleBooks Cloud without Enclave, Luca uses TripleBooks’s shared inference path: multi-tenant application isolation still applies (per-workspace databases, encryption, RBAC), while model serving is on TripleBooks-managed capacity shared across customers under TripleBooks’s provider contracts.
Enclave keeps that same application boundary and moves model serving into dedicated or on-prem capacity so inference traffic and retention policies match your enterprise controls.
Compared with pasting into ChatGPT or consumer MCP clients
Luca with Enclave keeps financial and operational context inside TripleBooks: tools run against your tenant APIs, actions stay permissioned, and inference stays on the Enclave runtime.
Copying balances, invoices, or documents into a consumer ChatGPT session — or wiring a personal MCP client to a general-purpose assistant outside TripleBooks — sends that material to those products’ endpoints and policies. Prefer Luca (and Enclave where required) when the work should remain on TripleBooks-controlled rails.
For API and IDE automation against your tenant, use MCP with entity-scoped API keys — that path talks to TripleBooks APIs directly; pair it with Enclave when you also need private in-product Luca inference.
Learn more
- Luca AI — conversational AI, agents, and guardrails
- Self-hosting — running the full application on your infrastructure
- Architecture — workspace isolation and encryption at a glance
- Glossary — Enclave and related terms