Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs: add AGENTS.md with machine-readable contribution rules (#322) (#325)
kuse386307bb1b9c
diff --git a/AGENTS.md b/AGENTS.mdnew file mode 100644index 00000000..cbc8a237--- /dev/null+++ b/AGENTS.md@@ -0,0 +1,40 @@+# AGENTS.md++Context for AI coding assistants (Claude Code, Codex, Cursor, Copilot, Gemini CLI) working in this repository.++**Laya** is a fast, local, on-device decision engine. These rules mirror [CONTRIBUTING.md](CONTRIBUTING.md); the assistant-facing summary lives here because coding agents read this file automatically.++## Do NOT++- Introduce a dependency on a hosted service or external API. Features must run in the user's own process or on their own hardware; a feature that only works against a hosted backend belongs in a separate integration, not here.+- Change the public API without updating `tests/test_hooks_api.py` in the same pull request — that suite is the API contract.+- Skip the CI gates before declaring a change done:++ ```bash+ ruff check laya/ --select=E9,F63,F7,F82,F401,F811 --line-length=120+ python -m compileall -q laya/ tests/+ ```++- Reformat files wholesale, reorder imports, or "modernize" surrounding code. Match the style of the file being edited; a diff full of formatting noise gets a change rejected.+- Commit secrets, tokens, or large binary files.+- Open pull requests in a language other than English. The project is triaged in English.+- Invent conventions. Use conventional commit prefixes matching the history (`feat(agent):`, `fix(router):`, `perf(common):`, `docs(hooks):`, `test(batch):`), and keep one logical change per commit.++## Where to look++| Editing | Read | Check |+|---|---|---|+| `laya/` core | — | plain script suites: `python tests/test_router.py`, `tests/test_criteria.py`, `tests/test_hooks.py`, `tests/test_hooks_api.py` |+| server path | `tests/test_serve.py` | `python -m pytest tests/test_serve.py` (skips when the `serve` extra is missing) |+| ONNX path | `tests/test_onnx.py` | `python -m pytest tests/test_onnx.py` (skips when the `onnx` extra is missing) |+| fast/local paths | `tests/test_fast.py`, `tests/test_local_e2e.py`, `tests/test_mcp_local_e2e.py` | need CUDA or checkpoints under `~/laya_models`; skipped otherwise |+| docs | `docs/` | — |++Optional extras are declared in `pyproject.toml` (`serve`, `fast`, `onnx`, `langchain`, `langgraph`); install only what the change needs.++## Pull requests++- Rebase onto the latest `main` so the diff is only your change.+- Keep it focused; split unrelated work into another PR.+- If a change moves numbers, report the before and after. The maintainer verifies decisions against real checkpoints, and measured deltas (probability changes, latency, memory) are what gets a change merged.+- Leave one runnable check behind for non-trivial logic; an assert-based script is enough.