Invite-only developer beta

Answers grounded in the memory your AI has already earned.

Longweis Relay is an OpenAI-compatible API for developers testing source-grounded answers over long-running AI memory.

Apply for one of 12 seats See the test →

Private beta · Text API · Non-sensitive test data only

relay-auto verified before return

Aug 20Order 001 is pending

Aug 21Order 001 was delivered

Latest stateOrder 001 was delivered.Source anchored to the Aug 21 event

A focused first cohort

Bring one real memory problem. Try to break the answer.

You build an agent, RAG system, assistant, or support workflow.
You can test with 20+ timestamped, non-sensitive events.
You can run three scenarios and send candid feedback within 10 days.

The beta protocol

Three tests. Ten days. Useful criticism.

Add test memory

Send timestamped, non-sensitive events through a tenant-bound memory endpoint.

Ask what changed

Use relay-auto through an OpenAI-compatible text client and test exact recall or chronology.

Grade the failure

Tell us what was wrong, missing, incompatible, or too slow. The sharp edges are the point.

A narrow technical contract

One key. One base URL. One verified response.

Relay currently supports a focused text subset of Chat Completions, plus immutable memory ingest and source-anchored context retrieval. It is not a tool-calling, multimodal, or production-SLA launch.

base_url = "https://api.longweisos.com/v1"
model    = "relay-auto"
key      = "displayed once after invitation"

12 private-beta seats

Tell us what you want to test.

We select for workflow fit, not company size. Accepted testers receive an individual one-time invitation and a separate test key.