Run the same checks on Linux that the Windows scripts run

The product could only reach a GPU through one hard-coded route: `wsl.exe -d
Ubuntu`, a device named "RTX 4090", and launch scripts that existed only as
PowerShell. On this machine `WSL_DISTRO_NAME` is already set, so the local
shell is Linux and the wsl.exe hop was unnecessary; on any other Linux host
it is impossible. Every SGLang, vLLM and KTransformers profile was reported
"missing" here even with a working 4090.

profiles.py now asks the host what it is instead of assuming: `uses_wsl()`
is false inside WSL and only true when wsl.exe is actually on PATH,
`host_shell()` runs `bash -lc` directly or through `wsl.exe -d
settings.wsl_distro`, `gpu_available()` replaced `wsl_gpu_available()`, and
`launch_script()` picks the .sh twin on a Linux host and the .ps1 on a
Windows one. The schema field `wsl_gpu` became `gpu_available` because the
old name lied on Linux.

Training no longer fails on a card name. `cuda_bf16_training()` gates on
cuda_available plus bf16_supported, which the training environment already
records, so an RTX 5090 or a 4080 passes and a non-BF16 4090 still fails.

Fourteen .sh twins cover the six launch profiles plus install, verify,
serve, download and finetune; the six WSL-hopping .ps1 files read
ELY_EYE_WSL_DISTRO instead of Ubuntu. PowerShell parsing is not verified
here because no pwsh exists on this host.

ci.yml runs pytest, ruff, go build/vet/test/gofmt, the Vite build and a
docker build-plus-curl on every push. To make that lint gate mean anything,
the 17 pre-existing ruff findings across backend/src are cleared: 14
mechanical fixes plus three probe handlers that now name the failure
families they absorb (OSError, SubprocessError, ValueError) instead of
catching Exception.

Verification: pytest 239 passed; ruff check backend/src bench clean (17 at
HEAD); mypy on the 12 touched files 14 errors, identical at HEAD; pyright on
the same files 8 errors, identical at HEAD; bash -n clean on all 15 .sh.

.github/workflows/ci.yml

@@ -0,0 +1,67 @@
+name: ci
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  backend:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+          cache: pip
+      - run: pip install -e ./backend
+      - run: python -m pytest -q
+      - run: pip install ruff==0.16.10 && ruff check backend/src bench
+
+  mcp:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version: "1.25"
+          cache-dependency-path: ely-eye-mcp/go.sum
+      - name: Build, vet and test
+        working-directory: ely-eye-mcp
+        run: go build ./... && go vet ./... && go test ./...
+      - name: gofmt is clean
+        working-directory: ely-eye-mcp
+        run: test -z "$(gofmt -l .)"
+
+  interface:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+          cache: npm
+          cache-dependency-path: frontend/package-lock.json
+      - run: npm ci && npm run build
+        working-directory: frontend
+
+  image:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Build the image from a clean clone
+        run: docker build -t ely-eye:ci .
+      - name: The container answers on both faces
+        run: |
+          docker run -d --name elyeye -p 8765:8765 ely-eye:ci
+          url=http://127.0.0.1:8765/api/status
+          for attempt in $(seq 1 30); do
+            curl -fsS "$url" -o /tmp/status.json && break
+            sleep 2
+          done
+          cat /tmp/status.json
+          curl -fsS http://127.0.0.1:8765/ | grep -q '<div id="root">'
+          docker logs elyeye

backend/src/ely_eye/api.py

@@ -277,8 +277,10 @@ def create_app() -> FastAPI:
         return benchmark_prediction_report(settings, settings.data_dir / "benchmark_predictions")
 
     @app.get("/api/runtime/profiles", response_model=RuntimeProfileReport)
-    def runtime_profiles() -> RuntimeProfileReport:
-        return runtime_profile_report()
+    def runtime_profiles(
+        settings: Annotated[Settings, Depends(settings_dep)],
+    ) -> RuntimeProfileReport:
+        return runtime_profile_report(settings)
 
     @app.get("/api/runtime/adapters", response_model=AdapterLoadReport)
     def runtime_adapters(settings: Annotated[Settings, Depends(settings_dep)]) -> AdapterLoadReport:

backend/src/ely_eye/cli.py

@@ -325,7 +325,7 @@ def run_benchmark(
 
 @app.command("runtime-profiles")
 def runtime_profiles() -> None:
-    typer.echo(runtime_profile_report().model_dump_json(indent=2))
+    typer.echo(runtime_profile_report(get_settings()).model_dump_json(indent=2))
 
 
 @app.command("research-modules")

backend/src/ely_eye/config.py

@@ -27,6 +27,7 @@ class Settings(BaseSettings):
     sglang_base_url: str = "http://127.0.0.1:8000/v1"
     sglang_api_key: str = "ely-eye-local"
     ui_dir: Path | None = None
+    wsl_distro: str = "Ubuntu"
     dense_embeddings_enabled: bool = False
     live_context_tokens: int = 262_144
     extreme_context_tokens: int = 1_010_000

backend/src/ely_eye/embeddings.py

@@ -2,7 +2,6 @@ from __future__ import annotations
 
 import json
 from dataclasses import dataclass
-from pathlib import Path
 from typing import Any
 
 import numpy as np

backend/src/ely_eye/evidence_graph.py

@@ -64,5 +64,5 @@ class EvidenceGraphService:
 
 
 def claim_node_id(question: str, answer: str) -> str:
-    digest = hashlib.sha256(f"{question}\n{answer}".encode("utf-8")).hexdigest()[:20]
+    digest = hashlib.sha256(f"{question}\n{answer}".encode()).hexdigest()[:20]
     return f"claim_{digest}"

backend/src/ely_eye/profiles.py

@@ -6,9 +6,9 @@ import shutil
 import subprocess
 from pathlib import Path
 
+from .config import Settings
 from .schemas import RuntimeLaunchProfile, RuntimeProfileReport, RuntimeProfileValidation
 
-
 EXTREME_YARN_OVERRIDE = {
     "text_config": {
         "rope_parameters": {
@@ -24,7 +24,7 @@ EXTREME_YARN_OVERRIDE = {
 }
 
 
-def runtime_profile_report(project_root: Path | None = None) -> RuntimeProfileReport:
+def runtime_profile_report(settings: Settings, project_root: Path | None = None) -> RuntimeProfileReport:
     root = project_root or Path.cwd()
     profiles = [
         RuntimeLaunchProfile(
@@ -185,29 +185,55 @@ def runtime_profile_report(project_root: Path | None = None) -> RuntimeProfileRe
             required_features=["GGUF", "offline demo", "quantized KV"],
         ),
     ]
-    validations = validate_runtime_profiles(root, profiles)
+    validations = validate_runtime_profiles(root, profiles, settings)
     ready = all(validation.status == "ready" for validation in validations)
     return RuntimeProfileReport(
         status="ready" if ready else "incomplete",
-        profiles=profiles,
+        profiles=[profile.model_copy(update={"script": launch_script(profile.script)}) for profile in profiles],
         validations=validations,
     )
 
 
+def uses_wsl() -> bool:
+    """A Windows process reaches Linux through wsl.exe; inside WSL the local shell already is Linux."""
+    if os.environ.get("WSL_DISTRO_NAME"):
+        return False
+    return shutil.which("wsl.exe") is not None
+
+
+def launch_script(script: str) -> str:
+    """A Windows host runs the .ps1; a Linux host, WSL included, runs the .sh twin."""
+    return script if uses_wsl() else str(Path(script).with_suffix(".sh"))
+
+
+def host_shell(command: str, settings: Settings) -> list[str]:
+    """Run a shell command where the launch environments live."""
+    if uses_wsl():
+        return ["wsl.exe", "-d", settings.wsl_distro, "bash", "-lc", command]
+    return ["bash", "-lc", command]
+
+
+def host_path(root: Path) -> str:
+    """The path the host shell can cd into; a Windows path is reached through the WSL mount."""
+    resolved = root.resolve()
+    return windows_path_to_wsl(resolved) if uses_wsl() else str(resolved)
+
+
 def validate_runtime_profiles(
     root: Path,
     profiles: list[RuntimeLaunchProfile],
+    settings: Settings,
 ) -> list[RuntimeProfileValidation]:
-    wsl_gpu = wsl_gpu_available()
+    gpu = gpu_available(settings)
     package_cache = {
-        ".venv-linux": linux_package_versions(root, ".venv-linux"),
-        ".venv-linux-vllm": linux_package_versions(root, ".venv-linux-vllm"),
+        ".venv-linux": linux_package_versions(root, ".venv-linux", settings),
+        ".venv-linux-vllm": linux_package_versions(root, ".venv-linux-vllm", settings),
     }
     probe_cache: dict[tuple[str, str, str | None], dict[str, object]] = {}
     llama_server = resolve_llama_server()
     validations: list[RuntimeProfileValidation] = []
     for profile in profiles:
-        script_exists = (root / profile.script).exists()
+        script_exists = (root / launch_script(profile.script)).exists()
         package = package_for_runtime(profile.runtime)
         env_name = linux_env_for_runtime(profile.runtime)
         linux_packages = package_cache.get(env_name, {})
@@ -216,24 +242,24 @@ def validate_runtime_profiles(
         executable = llama_server if profile.runtime == "llama.cpp" else None
         probe_key = (profile.runtime, env_name, executable)
         if probe_key not in probe_cache:
-            probe_cache[probe_key] = runtime_launcher_probe(root, profile.runtime, env_name, executable)
+            probe_cache[probe_key] = runtime_launcher_probe(root, profile.runtime, env_name, executable, settings)
         probe = probe_cache[probe_key]
         probe_status = str(probe.get("status", "skipped"))
         probe_passed = probe_status == "passed"
         if profile.runtime in {"SGLang", "vLLM", "KTransformers"}:
-            ready = script_exists and bool(package_version) and wsl_gpu and probe_passed
+            ready = script_exists and bool(package_version) and gpu and probe_passed
             if profile.runtime == "KTransformers":
                 ready = ready and bool(kt_sglang_version)
             detail = (
                 (
-                    f"WSL GPU ready; {package} {package_version}; "
+                    f"GPU ready; {package} {package_version}; "
                     f"sglang-kt entry {kt_sglang_version}"
                 )
                 if profile.runtime == "KTransformers" and ready
-                else f"WSL GPU ready; {package} {package_version}; env {env_name}"
+                else f"GPU ready; {package} {package_version}; env {env_name}"
                 if ready
                 else (
-                    f"script={script_exists}, wsl_gpu={wsl_gpu}, env={env_name}, "
+                    f"script={script_exists}, gpu={gpu}, env={env_name}, "
                     f"{package}={package_version}, sglang={kt_sglang_version}"
                 )
             )
@@ -253,7 +279,7 @@ def validate_runtime_profiles(
                 package=package,
                 package_version=package_version,
                 executable=executable,
-                wsl_gpu=wsl_gpu,
+                gpu_available=gpu,
                 probe_command=[str(item) for item in probe.get("command", [])],
                 probe_status=probe_status,
                 probe_exit_code=probe.get("exit_code") if isinstance(probe.get("exit_code"), int) else None,
@@ -278,7 +304,7 @@ def linux_env_for_runtime(runtime: str) -> str:
     return ".venv-linux"
 
 
-def linux_package_versions(root: Path, env_name: str) -> dict[str, str]:
+def linux_package_versions(root: Path, env_name: str, settings: Settings) -> dict[str, str]:
     script = (
         "import importlib.metadata as md, json;"
         "names=['sglang','vllm','ktransformers','torch','torchvision','qwen-vl-utils'];"
@@ -290,25 +316,21 @@ def linux_package_versions(root: Path, env_name: str) -> dict[str, str]:
     )
     try:
         output = subprocess.check_output(
-            [
-                "wsl.exe",
-                "-d",
-                "Ubuntu",
-                "bash",
-                "-lc",
-                (
-                    f"cd {wsl_quote(windows_path_to_wsl(root))} "
-                    f"&& test -x {env_name}/bin/python "
-                    f"&& {env_name}/bin/python -c {wsl_quote(script)}"
-                ),
-            ],
+            host_shell(
+                f"cd {wsl_quote(host_path(root))} "
+                f"&& test -x {env_name}/bin/python "
+                f"&& {env_name}/bin/python -c {wsl_quote(script)}",
+                settings,
+            ),
             text=True,
             stderr=subprocess.DEVNULL,
             timeout=20,
         )
-        data = json.loads(output.strip())
-        return {str(key): str(value) for key, value in data.items()}
-    except Exception:
+        payload = json.loads(output.strip())
+        if not isinstance(payload, dict):
+            return {}
+        return {str(key): str(value) for key, value in payload.items()}
+    except (OSError, subprocess.SubprocessError, ValueError):
         return {}
 
 
@@ -317,6 +339,7 @@ def runtime_launcher_probe(
     runtime: str,
     env_name: str,
     executable: str | None,
+    settings: Settings,
 ) -> dict[str, object]:
     if runtime == "SGLang":
         probe_script = (
@@ -325,7 +348,7 @@ def runtime_launcher_probe(
             "print(spec.origin if spec else 'missing'); "
             "sys.exit(0 if spec else 1)"
         )
-        command = wsl_python_command(root, env_name, f"-c {wsl_quote(probe_script)}")
+        command = host_python_command(root, env_name, f"-c {wsl_quote(probe_script)}", settings)
     elif runtime == "vLLM":
         probe_script = (
             "import importlib.util, sys; "
@@ -333,10 +356,10 @@ def runtime_launcher_probe(
             "print(spec.origin if spec else 'missing'); "
             "sys.exit(0 if spec else 1)"
         )
-        command = wsl_python_command(root, env_name, f"-c {wsl_quote(probe_script)}")
+        command = host_python_command(root, env_name, f"-c {wsl_quote(probe_script)}", settings)
     elif runtime == "KTransformers":
         probe_script = "import ktransformers, sglang; print('ktransformers and sglang import ok')"
-        command = wsl_python_command(root, env_name, f"-c {wsl_quote(probe_script)}")
+        command = host_python_command(root, env_name, f"-c {wsl_quote(probe_script)}", settings)
     elif runtime == "llama.cpp" and executable:
         command = [executable, "--help"]
     else:
@@ -349,7 +372,7 @@ def runtime_launcher_probe(
             timeout=30,
             check=False,
         )
-    except Exception as exc:
+    except (OSError, subprocess.SubprocessError) as exc:
         return {
             "status": "failed",
             "command": command,
@@ -365,13 +388,13 @@ def runtime_launcher_probe(
     }
 
 
-def wsl_python_command(root: Path, env_name: str, python_args: str) -> list[str]:
+def host_python_command(root: Path, env_name: str, python_args: str, settings: Settings) -> list[str]:
     shell_command = (
-        f"cd {wsl_quote(windows_path_to_wsl(root))} "
+        f"cd {wsl_quote(host_path(root))} "
         f"&& test -x {env_name}/bin/python "
         f"&& {env_name}/bin/python {python_args}"
     )
-    return ["wsl.exe", "-d", "Ubuntu", "bash", "-lc", shell_command]
+    return host_shell(shell_command, settings)
 
 
 def compact_output(stdout: str, stderr: str, limit: int = 800) -> str:
@@ -379,18 +402,17 @@ def compact_output(stdout: str, stderr: str, limit: int = 800) -> str:
     return text[:limit]
 
 
-def wsl_gpu_available() -> bool:
-    if shutil.which("wsl.exe") is None:
-        return False
+def gpu_available(settings: Settings) -> bool:
+    """Ask the host for a CUDA-capable GPU without naming a card: any GPU the driver lists will do."""
     try:
         output = subprocess.check_output(
-            ["wsl.exe", "-d", "Ubuntu", "nvidia-smi", "-L"],
+            host_shell("nvidia-smi -L", settings),
             text=True,
             stderr=subprocess.DEVNULL,
             timeout=10,
         )
         return "GPU" in output
-    except Exception:
+    except (OSError, subprocess.SubprocessError):
         return False
 
 

backend/src/ely_eye/proofs/checks_runtime.py

@@ -232,7 +232,7 @@ class RuntimeChecks:
             runtime.unload()
 
     def _runtime_profile_check(self, proof_root: Path) -> ProofCheck:
-        report = runtime_profile_report()
+        report = runtime_profile_report(self.settings)
         artifact = proof_root / "runtime_profiles.json"
         artifact.write_text(report.model_dump_json(indent=2), encoding="utf-8")
         required_runtimes = {"SGLang", "vLLM", "KTransformers", "llama.cpp"}

backend/src/ely_eye/proofs/training_trace.py

@@ -24,6 +24,11 @@ from ..training import (
 )
 
 
+def cuda_bf16_training(environment: dict[str, Any]) -> bool:
+    """The requirement is CUDA plus BF16 support, not one card name written down in 2025."""
+    return bool(environment.get("cuda_available")) and bool(environment.get("bf16_supported"))
+
+
 def count_safetensor_keys(path: Path) -> int:
     return int(safetensor_weight_stats(path)["tensor_count"])
 
@@ -173,8 +178,7 @@ def validate_training_trace(manifest: AdapterManifest, adapter_root: Path) -> di
         }
     )
     environment = summary.get("environment") if isinstance(summary.get("environment"), dict) else {}
-    device_name = str(environment.get("device_name") or "")
-    report["cuda_training"] = bool(environment.get("cuda_available")) and "RTX 4090" in device_name
+    report["cuda_training"] = cuda_bf16_training(environment)
     report["bf16_supported"] = bool(environment.get("bf16_supported"))
     report["valid"] = (
         report["trace_sha256_ok"]
@@ -282,11 +286,7 @@ def validate_training_proof(manifest: AdapterManifest, adapter_root: Path) -> di
                 and int(proof_weight_stats.get("nonzero_elements") or 0) > 0
                 and bool(proof_weight_stats.get("all_finite"))
             ),
-            "proof_cuda_bf16": (
-                bool(proof.get("bf16"))
-                and bool(proof_environment.get("cuda_available"))
-                and "RTX 4090" in str(proof_environment.get("device_name") or "")
-            ),
+            "proof_cuda_bf16": bool(proof.get("bf16")) and cuda_bf16_training(proof_environment),
             "proof_bf16_supported": bool(proof_environment.get("bf16_supported")),
             "proof_contract_ok": training_payload_contract_ok(proof, manifest),
             "proof_optimizer_ok": proof.get("optimizer") == "paged_adamw_8bit",

backend/src/ely_eye/research_modules.py

@@ -155,7 +155,7 @@ class ResearchModuleService:
                     "entries": layer.entries,
                     "hit_rate": hit_rate,
                     "evict": hit_rate < 0.25 and layer.entries > 0,
-                    "replay_token": hashlib.sha256(f"{layer.layer}:{layer.entries}".encode("utf-8")).hexdigest()[:16],
+                    "replay_token": hashlib.sha256(f"{layer.layer}:{layer.entries}".encode()).hexdigest()[:16],
                 }
             )
         return {"policy": "hit-rate weighted replay buffer", "layers": rows}

backend/src/ely_eye/schemas.py

@@ -439,7 +439,7 @@ class RuntimeProfileValidation(BaseModel):
     package: str | None = None
     package_version: str | None = None
     executable: str | None = None
-    wsl_gpu: bool = False
+    gpu_available: bool = False
     probe_command: list[str] = Field(default_factory=list)
     probe_status: Literal["passed", "failed", "skipped"] = "skipped"
     probe_exit_code: int | None = None

backend/src/ely_eye/training.py

@@ -9,7 +9,7 @@ import shutil
 import subprocess
 import time
 from contextlib import contextmanager
-from datetime import datetime, timezone
+from datetime import UTC, datetime
 from importlib.metadata import PackageNotFoundError, version
 from pathlib import Path
 from typing import Any, TextIO
@@ -20,7 +20,6 @@ from .config import Settings, get_settings
 from .db import Database
 from .schemas import AdapterManifest, EvidenceAtom, TrainingAdapterKind
 
-
 SYSTEM_PROMPT = (
     "You are Ely-Eye. Answer from supplied evidence, return compact JSON, "
     "and cite atom ids exactly."
@@ -433,7 +432,7 @@ class TrainingService:
             gradient_accumulation_steps or self.settings.train_gradient_accumulation_steps
         )
         learning_rate = learning_rate or self.settings.train_learning_rate
-        started_at = datetime.now(timezone.utc)
+        started_at = datetime.now(UTC)
         wall_start = time.perf_counter()
         trace_path = output_dir / "training_trace.jsonl"
 
@@ -575,7 +574,7 @@ class TrainingService:
         processor.save_pretrained(output_dir)
         self.write_lora_config(output_dir, kind, rank, alpha, model_id)
         weight_stats = safetensor_weight_stats(output_dir / "adapter_model.safetensors")
-        finished_at = datetime.now(timezone.utc)
+        finished_at = datetime.now(UTC)
         wall_seconds = time.perf_counter() - wall_start
         summary_path = output_dir / "training_summary.json"
         proof_path = output_dir / "training_proof.json"
@@ -671,7 +670,7 @@ class TrainingService:
         manifest = AdapterManifest(
             adapter_id=adapter_id(kind, dataset_path, model_id),
             kind=kind,
-            created_at=datetime.now(timezone.utc),
+            created_at=datetime.now(UTC),
             proof_version=TRAINING_PROOF_VERSION,
             training_method=TRAINING_METHOD,
             precision=TRAINING_PRECISION,
@@ -756,7 +755,7 @@ class TrainingService:
         output_dir.mkdir(parents=True, exist_ok=True)
         max_length = max_length or self.settings.train_max_length
         learning_rate = learning_rate or self.settings.train_learning_rate
-        started_at = datetime.now(timezone.utc)
+        started_at = datetime.now(UTC)
         wall_start = time.perf_counter()
         trace_path = output_dir / "training_trace.jsonl"
 
@@ -882,7 +881,7 @@ class TrainingService:
         processor.save_pretrained(output_dir)
         self.write_lora_config(output_dir, TrainingAdapterKind.hme_retrieval, rank, alpha, model_id)
         weight_stats = safetensor_weight_stats(output_dir / "adapter_model.safetensors")
-        finished_at = datetime.now(timezone.utc)
+        finished_at = datetime.now(UTC)
         wall_seconds = time.perf_counter() - wall_start
         summary_path = output_dir / "training_summary.json"
         proof_path = output_dir / "training_proof.json"
@@ -977,7 +976,7 @@ class TrainingService:
         manifest = AdapterManifest(
             adapter_id=adapter_id(TrainingAdapterKind.hme_retrieval, dataset_path, model_id),
             kind=TrainingAdapterKind.hme_retrieval,
-            created_at=datetime.now(timezone.utc),
+            created_at=datetime.now(UTC),
             proof_version=TRAINING_PROOF_VERSION,
             training_method=TRAINING_METHOD,
             precision=TRAINING_PRECISION,
@@ -1269,7 +1268,7 @@ def trainable_update_l2(before_update: dict[str, Any], named_parameters: list[tu
 
 def require_cuda_bf16(torch_module: Any) -> None:
     if not torch_module.cuda.is_available():
-        raise RuntimeError("PRD 12.2 training requires CUDA BF16 on the local RTX 4090.")
+        raise RuntimeError("training needs a CUDA device; none is available.")
     bf16_supported = getattr(torch_module.cuda, "is_bf16_supported", lambda: False)
     if not bool(bf16_supported()):
         raise RuntimeError("PRD 12.2 training requires BF16-capable CUDA hardware.")
@@ -1598,7 +1597,7 @@ def slugify(value: str) -> str:
 
 
 def adapter_id(kind: TrainingAdapterKind, dataset_path: Path, model_id: str) -> str:
-    raw = f"{kind.value}:{model_id}:{dataset_path.resolve()}:{datetime.now(timezone.utc).isoformat()}"
+    raw = f"{kind.value}:{model_id}:{dataset_path.resolve()}:{datetime.now(UTC).isoformat()}"
     return f"{kind.value}_{sha256_bytes(raw.encode('utf-8'))[:12]}"
 
 
@@ -1632,7 +1631,7 @@ def sha256_bytes(data: bytes) -> str:
 
 
 class _DigestWriter:
-    def __init__(self, digest: "hashlib._Hash") -> None:
+    def __init__(self, digest: hashlib._Hash) -> None:
         self.digest = digest
 
     def write(self, data: bytes) -> int:

backend/src/ely_eye/verifier.py

@@ -7,7 +7,6 @@ from typing import Any
 
 from .schemas import CitationReport, CompiledContext
 
-
 ATOM_PATTERN = re.compile(r"[A-Za-z0-9_./\\:-]+:(?:text|image|page|frame|region)[A-Za-z0-9_./\\:#=-]*")
 
 

backend/tests/test_host_profiles.py

@@ -0,0 +1,102 @@
+"""The launch layer must describe the machine it runs on, not the one it was written on."""
+
+from __future__ import annotations
+
+import os
+import stat
+from pathlib import Path
+
+import pytest
+
+from ely_eye import profiles
+from ely_eye.config import Settings
+from ely_eye.proofs.training_trace import cuda_bf16_training
+from ely_eye.training import require_cuda_bf16
+
+
+def _stub_nvidia_smi(tmp_path: Path, monkeypatch, output: str, exit_code: int = 0) -> None:
+    bin_dir = tmp_path / "bin"
+    bin_dir.mkdir()
+    stub = bin_dir / "nvidia-smi"
+    stub.write_text(f"#!/bin/sh\nprintf '%s\\n' '{output}'\nexit {exit_code}\n", encoding="utf-8")
+    stub.chmod(stub.stat().st_mode | stat.S_IEXEC)
+    monkeypatch.setenv("PATH", f"{bin_dir}{os.pathsep}{os.environ['PATH']}")
+
+
+def test_a_gpu_that_is_not_the_written_down_card_still_counts(tmp_path: Path, monkeypatch) -> None:
+    _stub_nvidia_smi(tmp_path, monkeypatch, "GPU 0: NVIDIA GeForce RTX 5090 (UUID: GPU-1234)")
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: False)
+    assert profiles.gpu_available(Settings(home=tmp_path / "home")) is True
+
+
+def test_a_machine_the_driver_cannot_see_reports_no_gpu(tmp_path: Path, monkeypatch) -> None:
+    _stub_nvidia_smi(tmp_path, monkeypatch, "nvidia-smi: command not found", exit_code=127)
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: False)
+    assert profiles.gpu_available(Settings(home=tmp_path / "home")) is False
+
+
+def test_the_distro_name_comes_from_settings(monkeypatch) -> None:
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: True)
+    command = profiles.host_shell("true", Settings(home=Path("/tmp"), wsl_distro="Ubuntu-24.04"))
+    assert command == ["wsl.exe", "-d", "Ubuntu-24.04", "bash", "-lc", "true"]
+
+
+def test_inside_wsl_the_local_shell_runs(monkeypatch, tmp_path: Path) -> None:
+    monkeypatch.setenv("WSL_DISTRO_NAME", "Ubuntu")
+    monkeypatch.setattr(profiles.shutil, "which", lambda name: "/mnt/c/WINDOWS/system32/wsl.exe")
+    assert profiles.uses_wsl() is False
+    monkeypatch.delenv("WSL_DISTRO_NAME")
+    assert profiles.uses_wsl() is True
+    assert profiles.host_shell("true", Settings(home=tmp_path / "home", wsl_distro="Ubuntu"))[0] == "wsl.exe"
+
+
+def test_a_linux_host_is_pointed_at_the_shell_twin(monkeypatch) -> None:
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: False)
+    assert profiles.launch_script("scripts/launch-sglang-live.ps1") == "scripts/launch-sglang-live.sh"
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: True)
+    assert profiles.launch_script("scripts/launch-sglang-live.ps1") == "scripts/launch-sglang-live.ps1"
+
+
+def test_the_training_verdict_does_not_name_a_card() -> None:
+    newer = {"cuda_available": True, "bf16_supported": True, "device_name": "NVIDIA GeForce RTX 5090"}
+    assert cuda_bf16_training(newer) is True
+    older = {"cuda_available": True, "bf16_supported": True, "device_name": "NVIDIA GeForce RTX 4080"}
+    assert cuda_bf16_training(older) is True
+    without_bf16 = {"cuda_available": True, "bf16_supported": False, "device_name": "NVIDIA GeForce RTX 4090"}
+    assert cuda_bf16_training(without_bf16) is False
+
+
+class _FakeCuda:
+    def __init__(self, available: bool, bf16: bool) -> None:
+        self._available = available
+        self._bf16 = bf16
+
+    def is_available(self) -> bool:
+        return self._available
+
+    def is_bf16_supported(self) -> bool:
+        return self._bf16
+
+
+class _FakeTorch:
+    def __init__(self, available: bool, bf16: bool) -> None:
+        self.cuda = _FakeCuda(available, bf16)
+
+
+def test_training_refuses_a_machine_without_cuda_by_capability_not_card() -> None:
+    with pytest.raises(RuntimeError) as refused:
+        require_cuda_bf16(_FakeTorch(available=False, bf16=False))
+    assert "4090" not in str(refused.value)
+    require_cuda_bf16(_FakeTorch(available=True, bf16=True))
+
+
+def test_every_profile_ships_a_script_for_this_host(tmp_path: Path, monkeypatch) -> None:
+    root = Path(__file__).resolve().parents[2]
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: False)
+    linux = profiles.runtime_profile_report(Settings(home=tmp_path / "home"), root)
+    assert [item.name for item in linux.validations if not item.script_exists] == []
+    assert all(item.script.endswith(".sh") for item in linux.profiles), linux.profiles
+    monkeypatch.setattr(profiles, "uses_wsl", lambda: True)
+    windows = profiles.runtime_profile_report(Settings(home=tmp_path / "home"), root)
+    assert [item.name for item in windows.validations if not item.script_exists] == []
+    assert all(item.script.endswith(".ps1") for item in windows.profiles), windows.profiles

scripts/download-models.sh

@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+for model in Qwen/Qwen3.5-9B Qwen/Qwen3.5-0.8B Qwen/Qwen3-VL-Embedding-8B; do
+  .venv/bin/python ./scripts/download_models.py --model "$model"
+done

scripts/install-cuda-torch.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+.venv/bin/python -m pip install --upgrade --no-cache-dir --index-url https://download.pytorch.org/whl/cu130 torch==2.12.0+cu130 torchvision==0.27.0+cu130
+.venv/bin/python -c 'import torch; print(torch.__version__); print(torch.cuda.is_available()); print(torch.version.cuda)'

scripts/install.sh

@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -d .venv ]; then
+  python3 -m venv .venv
+fi
+.venv/bin/python -m pip install --upgrade pip setuptools wheel
+.venv/bin/python -m pip install -e ./backend
+.venv/bin/python -m pip install --upgrade --no-cache-dir unsloth unsloth_zoo
+.venv/bin/python -m pip install --upgrade --no-cache-dir --index-url https://download.pytorch.org/whl/cu130 torch==2.12.0+cu130 torchvision==0.27.0+cu130
+.venv/bin/python -m pip install --upgrade --no-cache-dir transformers==5.9.0 datasets==4.8.5 dill==0.4.1 multiprocess==0.70.19 fsspec==2026.2.0 safetensors==0.8.0rc0
+(cd frontend && npm install)
+[ -f .env ] || cp .env.example .env
+.venv/bin/ely-eye status

scripts/launch-ktransformers-extreme.ps1

@@ -1,12 +1,13 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for the KTransformers Linux runtime."
 }
 
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
 $command = @"
 set -euo pipefail
 cd "$wslRoot"
@@ -20,4 +21,4 @@ python -m sglang.launch_server \
   --cpu-offload-gb 48
 "@
 
-wsl.exe -d Ubuntu -- bash -lc $command
+wsl.exe -d $distro -- bash -lc $command

scripts/launch-ktransformers-extreme.sh

@@ -0,0 +1,15 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -f .venv-linux/bin/activate ]; then
+  echo "Run scripts/setup-linux-runtimes.sh before launching SGLang." >&2
+  exit 1
+fi
+source .venv-linux/bin/activate
+export SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1
+exec python -m sglang.launch_server \
+  --model-path Qwen/Qwen3.5-9B \
+  --port 8002 \
+  --context-length 1010000 \
+  --kt-cpuinfer 16 \
+  --cpu-offload-gb 48

scripts/launch-llamacpp-offline.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ -z "${ELY_EYE_GGUF_MODEL:-}" ]; then
+  echo "Set ELY_EYE_GGUF_MODEL to the absolute GGUF model path." >&2
+  exit 1
+fi
+server="${LLAMA_CPP_SERVER:-llama-server}"
+if ! command -v "$server" >/dev/null 2>&1; then
+  echo "llama-server executable is missing. Build llama.cpp or set LLAMA_CPP_SERVER." >&2
+  exit 1
+fi
+exec "$server" \
+  --model "$ELY_EYE_GGUF_MODEL" \
+  --host 127.0.0.1 \
+  --port 8003 \
+  --ctx-size 262144 \
+  --flash-attn \
+  --cache-type-k q4_0 \
+  --cache-type-v q4_0

scripts/launch-sglang-extreme.ps1

@@ -1,13 +1,14 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for the SGLang Linux runtime."
 }
 
 $override = '{"text_config":{"rope_parameters":{"mrope_interleaved":true,"mrope_section":[11,11,10],"rope_type":"yarn","rope_theta":10000000,"partial_rotary_factor":0.25,"factor":4.0,"original_max_position_embeddings":262144}}}'
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
 $command = @"
 set -euo pipefail
 cd "$wslRoot"
@@ -22,4 +23,4 @@ python -m sglang.launch_server \
   --json-model-override-args '$override'
 "@
 
-wsl.exe -d Ubuntu -- bash -lc $command
+wsl.exe -d $distro -- bash -lc $command

scripts/launch-sglang-extreme.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -f .venv-linux/bin/activate ]; then
+  echo "Run scripts/setup-linux-runtimes.sh before launching SGLang." >&2
+  exit 1
+fi
+source .venv-linux/bin/activate
+export SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN=1
+exec python -m sglang.launch_server \
+  --model-path Qwen/Qwen3.5-9B \
+  --port 8000 \
+  --tp-size 1 \
+  --mem-fraction-static 0.88 \
+  --context-length 1010000 \
+  --json-model-override-args '{"text_config":{"rope_parameters":{"mrope_interleaved":true,"mrope_section":[11,11,10],"rope_type":"yarn","rope_theta":10000000,"partial_rotary_factor":0.25,"factor":4.0,"original_max_position_embeddings":262144}}}'

scripts/launch-sglang-live.ps1

@@ -1,12 +1,13 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for the SGLang Linux runtime."
 }
 
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
 $command = @"
 set -euo pipefail
 cd "$wslRoot"
@@ -23,4 +24,4 @@ python -m sglang.launch_server \
   --speculative-num-draft-tokens 4
 "@
 
-wsl.exe -d Ubuntu -- bash -lc $command
+wsl.exe -d $distro -- bash -lc $command

scripts/launch-sglang-live.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -f .venv-linux/bin/activate ]; then
+  echo "Run scripts/setup-linux-runtimes.sh before launching SGLang." >&2
+  exit 1
+fi
+source .venv-linux/bin/activate
+exec python -m sglang.launch_server \
+  --model-path Qwen/Qwen3.5-9B \
+  --port 8000 \
+  --tp-size 1 \
+  --mem-fraction-static 0.82 \
+  --context-length 262144 \
+  --reasoning-parser qwen3 \
+  --speculative-algo NEXTN \
+  --speculative-num-steps 3 \
+  --speculative-num-draft-tokens 4

scripts/launch-vllm-extreme.ps1

@@ -1,12 +1,13 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for the vLLM Linux runtime."
 }
 
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
 $ropeScaling = '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":262144}'
 $command = @"
 set -euo pipefail
@@ -23,4 +24,4 @@ python -m vllm.entrypoints.openai.api_server \
   --rope-scaling '$ropeScaling'
 "@
 
-wsl.exe -d Ubuntu -- bash -lc $command
+wsl.exe -d $distro -- bash -lc $command

scripts/launch-vllm-extreme.sh

@@ -0,0 +1,17 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -f .venv-linux-vllm/bin/activate ]; then
+  echo "Run scripts/setup-linux-runtimes.sh before launching vLLM." >&2
+  exit 1
+fi
+source .venv-linux-vllm/bin/activate
+exec python -m vllm.entrypoints.openai.api_server \
+  --model Qwen/Qwen3.5-9B \
+  --host 127.0.0.1 \
+  --port 8001 \
+  --max-model-len 1010000 \
+  --enable-prefix-caching \
+  --gpu-memory-utilization 0.88 \
+  --trust-remote-code \
+  --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":262144}'

scripts/launch-vllm-live.ps1

@@ -1,12 +1,13 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for the vLLM Linux runtime."
 }
 
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
 $command = @"
 set -euo pipefail
 cd "$wslRoot"
@@ -21,4 +22,4 @@ python -m vllm.entrypoints.openai.api_server \
   --trust-remote-code
 "@
 
-wsl.exe -d Ubuntu -- bash -lc $command
+wsl.exe -d $distro -- bash -lc $command

scripts/launch-vllm-live.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if [ ! -f .venv-linux-vllm/bin/activate ]; then
+  echo "Run scripts/setup-linux-runtimes.sh before launching vLLM." >&2
+  exit 1
+fi
+source .venv-linux-vllm/bin/activate
+exec python -m vllm.entrypoints.openai.api_server \
+  --model Qwen/Qwen3.5-9B \
+  --host 127.0.0.1 \
+  --port 8001 \
+  --max-model-len 262144 \
+  --enable-prefix-caching \
+  --gpu-memory-utilization 0.82 \
+  --trust-remote-code

scripts/run-prd-finetune.sh

@@ -0,0 +1,35 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+max_steps="${MAX_STEPS:-37}"
+rank="${RANK:-16}"
+alpha="${ALPHA:-32}"
+max_length="${MAX_LENGTH:-8192}"
+gradient_accumulation_steps="${GRADIENT_ACCUMULATION_STEPS:-4}"
+cartridge_id="${CARTRIDGE_ID:-}"
+
+dataset=.ely_eye/training/hme_prd_sft.jsonl
+adapter=.ely_eye/adapters/hme_core_lora_qwen35_9b
+
+.venv/bin/ely-eye build-prd-training-data \
+  --prd-path ./PRD.md \
+  --output-path "$dataset" \
+  --max-sections 48
+
+args=(train-lora "$dataset" "$adapter"
+  --max-steps "$max_steps"
+  --kind hme_core_lora
+  --rank "$rank"
+  --alpha "$alpha"
+  --max-length "$max_length"
+  --gradient-accumulation-steps "$gradient_accumulation_steps")
+
+if [ -n "$cartridge_id" ]; then
+  args+=(--cartridge-id "$cartridge_id")
+fi
+
+.venv/bin/ely-eye "${args[@]}"
+
+echo "Adapter ready: $adapter"
+echo "Use with: ELY_EYE_ADAPTER_DIR=$adapter"

scripts/serve-backend.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+exec .venv/bin/ely-eye serve --host 127.0.0.1 --port 8765

scripts/serve-frontend.sh

@@ -0,0 +1,5 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+cd frontend
+exec npm run dev

scripts/setup-linux-runtimes.ps1

@@ -1,20 +1,21 @@
 $ErrorActionPreference = "Stop"
 
 Set-Location (Split-Path -Parent $PSScriptRoot)
+$distro = if ($env:ELY_EYE_WSL_DISTRO) { $env:ELY_EYE_WSL_DISTRO } else { "Ubuntu" }
 
 if (-not (Get-Command wsl.exe -ErrorAction SilentlyContinue)) {
   throw "WSL is required for SGLang, vLLM, and KTransformers runtime profiles."
 }
 
-wsl.exe -d Ubuntu -- bash -lc "command -v nvidia-smi >/dev/null && nvidia-smi -L >/dev/null"
-wsl.exe -d Ubuntu -- bash -lc "command -v curl >/dev/null || (apt-get update && apt-get install -y curl)"
-wsl.exe -d Ubuntu -- bash -lc "command -v /root/.local/bin/uv >/dev/null || curl -LsSf https://astral.sh/uv/install.sh | sh"
-$wslRoot = (wsl.exe -d Ubuntu wslpath -a (Get-Location).Path).Trim()
-wsl.exe -d Ubuntu -- bash -lc "cd '$wslRoot' && /root/.local/bin/uv venv .venv-linux --python 3.12"
-wsl.exe -d Ubuntu -- bash -lc "cd '$wslRoot' && /root/.local/bin/uv venv .venv-linux-vllm --python 3.12"
-wsl.exe -d Ubuntu -- bash -lc "cd '$wslRoot' && source .venv-linux/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-sglang-linux.txt"
-wsl.exe -d Ubuntu -- bash -lc "cd '$wslRoot' && source .venv-linux-vllm/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-vllm-linux.txt"
-wsl.exe -d Ubuntu -- bash -lc "cd '$wslRoot' && source .venv-linux/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-ktransformers-linux.txt"
+wsl.exe -d $distro -- bash -lc "command -v nvidia-smi >/dev/null && nvidia-smi -L >/dev/null"
+wsl.exe -d $distro -- bash -lc "command -v curl >/dev/null || (apt-get update && apt-get install -y curl)"
+wsl.exe -d $distro -- bash -lc "command -v /root/.local/bin/uv >/dev/null || curl -LsSf https://astral.sh/uv/install.sh | sh"
+$wslRoot = (wsl.exe -d $distro wslpath -a (Get-Location).Path).Trim()
+wsl.exe -d $distro -- bash -lc "cd '$wslRoot' && /root/.local/bin/uv venv .venv-linux --python 3.12"
+wsl.exe -d $distro -- bash -lc "cd '$wslRoot' && /root/.local/bin/uv venv .venv-linux-vllm --python 3.12"
+wsl.exe -d $distro -- bash -lc "cd '$wslRoot' && source .venv-linux/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-sglang-linux.txt"
+wsl.exe -d $distro -- bash -lc "cd '$wslRoot' && source .venv-linux-vllm/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-vllm-linux.txt"
+wsl.exe -d $distro -- bash -lc "cd '$wslRoot' && source .venv-linux/bin/activate && /root/.local/bin/uv pip install --prerelease=allow -r backend/requirements-ktransformers-linux.txt"
 
 if (-not (Get-Command llama-server -ErrorAction SilentlyContinue)) {
   winget install --id ggml.llamacpp -e --accept-package-agreements --accept-source-agreements

scripts/setup-linux-runtimes.sh

@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+if ! command -v nvidia-smi >/dev/null 2>&1; then
+  echo "nvidia-smi is missing; these runtimes need a CUDA host." >&2
+  exit 1
+fi
+nvidia-smi -L >/dev/null
+uv_bin="${ELY_EYE_UV:-$HOME/.local/bin/uv}"
+if ! command -v "$uv_bin" >/dev/null 2>&1; then
+  curl -LsSf https://astral.sh/uv/install.sh | sh
+  uv_bin="$HOME/.local/bin/uv"
+fi
+for venv in .venv-linux .venv-linux-vllm; do
+  "$uv_bin" venv "$venv" --python 3.12
+done
+"$uv_bin" pip install --python .venv-linux/bin/python --prerelease=allow -r backend/requirements-sglang-linux.txt
+"$uv_bin" pip install --python .venv-linux-vllm/bin/python --prerelease=allow -r backend/requirements-vllm-linux.txt
+"$uv_bin" pip install --python .venv-linux/bin/python --prerelease=allow -r backend/requirements-ktransformers-linux.txt
+.venv/bin/ely-eye runtime-profiles

scripts/verify.sh

@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+set -euo pipefail
+cd "$(dirname "$0")/.."
+.venv/bin/python -m compileall ./backend/src
+.venv/bin/ely-eye status
+(cd frontend && npm run build)