fix(router): lazy-load onnxruntime-node to stop crashing Windows imports - #230
Merged
Merged
Conversation
onnx-local.ts and onnx-local-optimized.ts both did a top-level
`await import('onnxruntime-node')` at module load time, which fires the
native-binding load (onnxruntime_binding.node) as a side effect of merely
importing the module — even for consumers that never touch the ONNX
router at all (e.g. reasoningbank, via core/distill -> router/router).
On Windows environments where the NAPI binary can't load, this crashes
the whole import chain.
Moves the load into a `loadOrt()` helper invoked lazily from
initializeSession() (onnx-local.ts), and drops the eager load entirely
from onnx-local-optimized.ts, which never referenced the resulting `ort`
binding at all — the base class it extends is the only thing that needs it.
Rebased from PR #155 (thanks to the original author) — cherry-picked just
this lazy-load refactor onto current main, since #155 was based on a
pre-#154/#228 snapshot and its @xenova/transformers hunks no longer apply
(main has since migrated fully to @huggingface/transformers).
Verified empirically (not just by reasoning about it): stubbed
onnxruntime-node with a load-marker side effect, confirmed the marker
fires on unconditional `import()` of the pre-fix module (bug reproduced)
and does not fire post-fix — while still firing once `chat()` actually
invokes initializeSession() (confirming lazy-load isn't just "never loads").
Closes #155.
Co-Authored-By: RuFlo <ruv@ruv.net>
Claude-Session: https://claude.ai/code/session_01SjxUvZnyBdwtx4Y1jKAqt6
ruvnet
added a commit
that referenced
this pull request
Sep 16, 2026
Patch release consolidating this cycle's PR triage (see ADR-078): - Windows fix: onnxruntime-node is now lazy-loaded instead of loaded at module-import time, which crashed environments where the native binding can't load (#230, rebased from #155). - ESM __dirname polyfill for agentdb-cli.ts (#78). - agentic-jujutsu: jj bookmark migration for jj >=0.21 (#174), darwin-arm64 optional dependency (#160). - Fully-local (no API key) path fixes: PROXY_PORT threading, --provider ollama exclusion, ONNX model path centralization (#228). - SECURITY.md stub (#166), 32 new example applications (#62). No breaking changes, no public API changes. Claude-Session: https://claude.ai/code/session_01SjxUvZnyBdwtx4Y1jKAqt6
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Rebases the core fix from #155 onto current main. `onnx-local.ts` and `onnx-local-optimized.ts` both did a top-level `await import('onnxruntime-node')` at module load time, firing the native-binding load as a side effect of merely importing the module — even for consumers that never touch the ONNX router (e.g. `reasoningbank`, via `core/distill -> router/router`). On Windows environments where the NAPI binary can't load, this crashes the whole import chain.
#155 couldn't be merged as-is: it was based on a pre-#154/#228 snapshot, and its `@xenova/transformers` hunks no longer apply now that main has fully migrated to `@huggingface/transformers`. This PR cherry-picks just the lazy-load refactor, preserving the `PHI4_MODEL_PATH`/`MODEL_ROOT` centralization #228 already landed.
Changes
Verification
Reasoned proof isn't enough for an import-time side effect, so this was verified empirically: stubbed `onnxruntime-node` with a load-marker side effect and confirmed:
`npx tsc -p config/tsconfig.json --skipLibCheck` — clean, no new errors.
Closes #155.
🤖 Generated with RuFlo
https://claude.ai/code/session_01SjxUvZnyBdwtx4Y1jKAqt6