DocketX / DocketRouter
General model routers solved a real problem: one API, many models, sane failover, usage in one place. DocketRouter is that same shape — an OpenAI-compatible endpoint, model fallbacks, per-request accounting, OAuth key issuance, bring-your-own-keys — built for the one domain where the expensive failure is not downtime or price. In legal work the expensive failure is a fluent answer citing a case that does not exist, and no general router has any opinion about that.
Same shape, different job
OpenAI-compatible chat completions, streaming, model fallbacks that advance on error and on a blank answer — a failure general routers miss because a confident empty 200 looks healthy at the edge.
Prepaid credits, per-key caps and jurisdiction defaults, per-generation cost and latency records, app attribution, OAuth PKCE key issuance for third-party apps.
Live per-model success and latency tracking; bring your own provider keys, encrypted at rest, or run private models on your own hardware.
The layer that is ours
The question retrieves controlling authority from 615,391 Texas decisions, 122,681 statutes, 45,660 Supreme Court documents and 6,093 verbatim rule sections — before the model writes.
Every citation checked against 18,123,788 citations; case names checked against the courts' records; rule quotations checked word for word.
What cannot be verified is stripped from the answer and named — even mid-stream, where the citation is held until its verdict lands.
That is the whole thesis: routing is a commodity, verification is not. A legal router without a library is a general router with a domain name.
Build on it
Third-party product names are used for comparison and identification only; no affiliation or endorsement is implied. Our figures are counted from the live library and independently checkable through the public API. Not legal advice.