- Root LICENSE: replace the short-form notice with the full canonical Apache-2.0 text (fixes GitHub's "Other" license detection); update the OpenJev -> Umpirix copyright line in LICENSE/NOTICE, otsone/NOTICE and sdk/python/LICENSE+NOTICE (missed by the rebrand pass). - sdk/ai-sdk: was MIT with no LICENSE/NOTICE and no upstream fork trail (unlike sdk/js, sdk/python, adapters/python, skills, which are real TypeSafe AI SDK forks and correctly keep MIT per their NOTICE). It is original Umpirix code, so switch it to Apache-2.0 with its own LICENSE/NOTICE, matching the root. - README Licensing section: list sdk/ai-sdk under Apache-2.0 and sdk/scribase-decisions under MIT, and spell out why the TypeSafe-forked directories stay MIT (their MIT terms require preserving the notice). - Audited caelum0x/openjev-private (read-only clone): it's an older, mostly-superseded snapshot of this repo. Migrated the 3 self-contained, secret-free example scripts it had that this repo didn't (raw-HTTP, no SDK dependency): content-moderation, lead-scoring, support-ticket-routing (Python + Node.js, each with a sample JSONL), plus scripts/reproduce-leaderboard.mjs (wired to a new `npm run bench:reproduce`) for leaderboard reproducibility. Left behind in private: src/lib/seo/retired.ts (a reversed decision - this repo now has live /compare pages) and the hosted-SaaS pieces already superseded by this repo's own, newer versions of the same files. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Examples
Runnable scripts against a running OpenJev server (npm run dev or
docker compose up). Each batches a typed question set (Choice + Score + Noul) per item
in one request, then applies a ranking/routing policy in plain code over the returned
probabilities — the model only judges the candidates you hand it, the decision logic stays
in your code and is free to change without another inference call.
| Example | Language | Shows |
|---|---|---|
rank-ads.mjs |
Node.js (built-in fetch, no dependency) |
Ranking ad-copy hooks by a composite score built from score + two noul flags |
triage-tickets.py |
Python (openjev SDK) |
Routing support tickets by department (choice) and a priority built from score + noul |
official-sdk.py |
Python (official typesafe-sdk from PyPI) |
The unmodified TypeSafe SDK against OpenJev: choice, score, noul, several questions at once, models.list(), a rejected request |
official-sdk.mjs |
Node.js (official @typesafe-ai/sdk from npm) |
The same, plus a zstd-compressed request body |
content-moderation/triage.py / .ts |
Python + Node.js (raw HTTP, no SDK) | Sorting comments into publish / hold / human-review from a choice category plus confidence, over posts.jsonl |
lead-scoring/score.py / .ts |
Python + Node.js (raw HTTP, no SDK) | Three typed questions per lead (stage score, segment choice, inquiry-vs-pitch choice) combined into a points formula in code, over leads.jsonl |
support-ticket-routing/route.py / .ts |
Python + Node.js (raw HTTP, no SDK) | Team choice + urgency score + a refund noul flag per ticket, auto-routed vs sent to a human queue, over tickets.jsonl |
The three above use only urllib/fetch against the plain HTTP API — no SDK install —
for anyone self-hosting without either language's client. triage-tickets.py shows the
same idea through the openjev SDK client instead.
Run either with the server already up:
npm run dev # http://localhost:8080
OPENJEV_BASE_URL=http://localhost:8080 node examples/rank-ads.mjs
pip install ./sdk/python && python examples/triage-tickets.py
The two official-SDK examples need their package installed first and read the SDK's own variables:
pip install typesafe-sdk && TYPESAFE_BASE_URL=http://localhost:8080 python examples/official-sdk.py
npm install @typesafe-ai/sdk && TYPESAFE_BASE_URL=http://localhost:8080 node examples/official-sdk.mjs
The first two read OPENJEV_BASE_URL (default http://localhost:3000 for the Node example,
http://localhost:8080 for the Python one — match whichever you started). No API key is
needed: the SDK sends no Authorization header unless OPENJEV_API_KEY is set, which you
only do when your server issues keys.
feedback.txt is sample input for the no-server command line classifier:
node scripts/cli.mjs classify bug,feature,praise < examples/feedback.txt (see the main README).
classify-eval.jsonl is a synthetic, hand-labeled gold set (65
rows, built for this repo, not scraped) for measuring openjev classify's lexical-engine
accuracy, including rows that deliberately reuse a different label's vocabulary
(shared_vocab: true). Run npm run classify-eval — see the "Lexical-engine accuracy"
section in the main README for the measured before/after numbers.