LLM proxies
How Chordian picks the model behind each memory — the R1 default, your own key, and sovereign boxes.
Every memory is powered by a large language model for reasoning, synthesis, and answering. Chordian lets you choose which one — while keeping embeddings and data sovereignty consistent.
The default: Chordian R1
Out of the box, a Chordian-hosted memory uses Chordian R1 — a self-hosted model running on Chordian infrastructure. It requires zero setup and no keys of your own. On a memory's AI Model & Keys page it appears as "Use Chordian — Powered by Chordian R1" and is selected by default.
R1 powers chat, synthesis, and contradiction detection. Knowledge-graph extraction and embeddings run on Chordian's embedding model regardless of the chat model you pick, so your search results and vector index stay consistent.
Bringing your own model
On the same AI Model & Keys page you can point a memory at your own OpenAI key instead. Your key is stored encrypted (in Chordian's Vault for hosted memories, or on your own box for BYOI), and usage on your key is billed to your provider account — not metered as Chordian credits.
Switching a single memory to your own model does not affect any other memory: the model is chosen per memory.
What powers each memory type
- Shared and Dedicated (CDMBx) memories — run on Chordian infrastructure and use Chordian R1 by default, with per-tenant isolation and data-perimeter filtering.
- Own-infrastructure (BYOI) memories — are sovereign and air-gapped. They route LLM traffic to their own configured endpoint (the Chordian proxy over the secure tunnel, your own OpenAI-compatible endpoint, or a model running inside the box). Chordian's default model is never forced on a sovereign box, and a box with no model configured refuses to send your content to a public provider rather than falling back silently.
Metering
When a memory uses Chordian R1, its token usage is metered in credits — see Credits and limits. Input and output tokens are priced per model, converted at the platform exchange rate, and shown on the memory's LLM Usage page. Usage on your own key is displayed for transparency but charged 0 credits.