"How do we know our data isn't used for training?"
Contractual no-training / no-secondary-use commitment (doc 16). Processing is limited to serving the request, guardrails, and the audit trail. Point to the trust page.
Sales enablement
Dolphin sales one-pager
For sales and CS. Internal positioning, objection handling, and competitive context. Customer-facing summaries must be reviewed before they leave this portal.
The one-liner
Dolphin is a neutral, multi-provider AI governance plane for Japanese mid-market enterprises. One SSO chat UI plus one OpenAI-compatible API endpoint, where every request is logged, DLP-screened, and rolled up into an exportable audit ledger (監査台帳) the customer can show in an ISMS / Pマーク audit.
Buyer & pain
- Buyer: 情シス / CISO at a Japanese mid-market enterprise.
- Pain: shadow AI — employees pasting confidential data into consumer AI tools with no visibility, no controls, and nothing to show an auditor.
- Wedge: shadow-AI consolidation & audit. Give them one governed front door and the evidence trail their auditor wants.
Why we win
- Neutrality: BYOK and/or our managed models — not locked to one vendor. Provider is hidden behind aliases at the surface.
- Japanese-compliance productization: the audit ledger maps directly to ISMS / Pマーク evidence; JP-sovereign provider options available.
- Managed convenience: they don't have to wire up providers themselves; we run the governance plane.
Objection handling
"Does our data stay in Japan?"
We do not claim japan_only in v1. Lead with JP-sovereign provider options (さくら, Azure Japan East) and the East-Tokyo posture for primary data + keys. Compute may be global in v1 — be honest; see the provider matrix.
"Why are some providers off by default?"
China-domiciled providers (DeepSeek, Kimi) are default-off and gated per tenant. Frame it as a control, not a limitation: we won't route to them without explicit approval.
"Isn't this just a proxy?"
No — it's the governance plane: SSO, per-model guardrails, DLP, budgets, and an immutable, exportable audit ledger. The OpenAI-compatible endpoint is one surface, not the product.
Competitive context
- vs OpenRouter / LiteLLM: those are routing layers; we add Japanese-compliance governance + audit evidence on top.
- vs ChatGPT Enterprise: single-vendor lock-in and no BYOK neutrality; weaker fit for a 情シス wanting provider-neutral control + JP audit artifacts.
Guardrail for customer-facing material
When you copy anything from this page into a customer deck or the help portal, strip internal provider names, routing details, and margin. Use model aliases only. See the provider matrix for what stays internal.