Skip to content

Company

Infrastructure, operated in public.

Why Caedral exists: one system for models, hosted products, billing, and optional request preparation — without fake headcount, offices, or investors.

Get API keyContact

One API

Chat, embeddings, rerank, and voice on a single OpenAI-compatible surface.

Caedral technology

Embed, Rerank, and Voice are Caedral-hosted products — not a thin wrapper around a single lab.

Notre

Included on every plan. Applied on eligible chat when the platform runs Notre. Fail-open. Not a billed SKU.

Billing

Monthly pools. HTTP 402 at zero. On-demand only when you enable it.

How we operate

Policies for teams evaluating Caedral as production infrastructure.

Caedral exists so production applications can use many models without assembling a pile of provider accounts. The company operates a gateway, hosted models, subscription billing, and Notre — a research program that may prepare chat requests before inference.

Mission

Make unified AI infrastructure boring in the right way: one key, real model IDs, published rates, a hard stop on spend, and honest maturity labels for work that is still experimental.

Product philosophy

  • Truth over theater — do not advertise roadmap work as production
  • Caedral-owned products deserve first-class pages, not footnotes in a catalog
  • External labs stay named; we do not hide who trained a model
  • Notre is opt-in and fail-open; it does not currently sit on every path
  • Developers should leave with a next step: docs, a model ID, or the dashboard

Unified AI infrastructure

The public surface is an OpenAI-compatible API, a live catalog, dual usage pools, teams billing, a dashboard, and documentation. Behind that: authentication, rate limits, ledgered usage, and routing to hosted or external inference.

Notre research

Notre is how Caedral studies request preparation — compiler-like work on conversations, tools, and redundant context. Engine and DB are in-development or pre-alpha. Production currently accepts the Notre field and strips it so clients can ship against the contract without depending on an optimizer that is not yet applied in production.

Uptime target

Paid plans target 99.9% monthly uptime; Hobby is best-effort and rate-limited. Live component status is published at /status. We do not invent latency benchmarks; measured figures appear on status when we have them.

Data privacy

We do not train on your prompts or completions by default. API traffic is processed to fulfill requests, enforce billing and rate limits, and operate the service. Execution logs may be retained for debugging and account support under documented retention. Deletion requests: dashboard or [email protected].

Support

Community and integration: Discord. Account and billing: [email protected]. Paying accounts receive priority handling on billing and production incidents.