v4.7.2

roboflow/supervisionv4.7.2Oct 2, 2026by github-actions[bot]

Upgrading?

AI Summary

A patch release improving Node.js runtime defaults, fixing parallel tool result ordering in chat agents, and resolving download failures for large payloads.

Key Highlights

  • Uses the current Node.js LTS for deployments where the runtime is unset or set to 'node'.
  • Chat agents now send parallel tool results in the order tools were called.
  • Recovering long unfinished chat responses no longer blocks the worker.
  • Fixed 'Download payload' and 'Download output' buttons for large data.

Full Release Notes

## Upgrade

```sh
npx trigger.dev@latest update  # npm
pnpm dlx trigger.dev@latest update  # pnpm
yarn dlx trigger.dev@latest update  # yarn
bunx trigger.dev@latest update  # bun
```

Self-hosted Docker image: [`ghcr.io/triggerdotdev/trigger.dev:v4.7.2`](https://github.com/triggerdotdev/trigger.dev/pkgs/container/trigger.dev/1327463716?tag=v4.7.2)

## Release notes

Read the full release notes: https://trigger.dev/changelog/v4-7-2

## What's changed

## Improvements
- Use the current Node.js LTS for deployments whose `runtime` is unset or set to `"node"`. Set a versioned runtime such as `"node-24"` to keep deployments pinned to that Node.js major. ([`9ecdf735e`](https://github.com/triggerdotdev/trigger.dev/commit/9ecdf735e7a3263bd4605e0e34c7c5484964a0a8))
- `chat.agent` now sends parallel tool results to the model in the order the tools were called, the same order the next turn's history uses. Models that bind their thinking to the exact conversation (Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1) no longer lose a turn's reasoning, or fail the request on accounts that enforce the check, when its tools finish out of order. ([`760737258`](https://github.com/triggerdotdev/trigger.dev/commit/76073725801988056d23bbdde76831d81e671cd7))
- Recovering a long unfinished `chat.agent` response on a continuation run no longer blocks the worker. Rebuilding the response from its streamed chunks now takes time linear in its length and yields to the event loop as it goes, so heartbeats keep firing and the run is not killed mid-replay. Capturing the response at the end of a long turn gets the same speedup. ([`93c63c685`](https://github.com/triggerdotdev/trigger.dev/commit/93c63c6859648521792bd4d2cd7c0c5966eef991))

## Server changes

These changes affect the self-hosted Docker image and Trigger.dev Cloud:

- Fix the "Download payload" and "Download output" buttons for large payloads and outputs, which downloaded an error file instead of the data.

## All packages: v4.7.2

@trigger.dev/build, @trigger.dev/core, @trigger.dev/python, @trigger.dev/react-hooks, @trigger.dev/redis-worker, @trigger.dev/rsc, @trigger.dev/schema-to-json, @trigger.dev/sdk, trigger.dev

## Contributors

Eric Allam, Chris Arderne, github-actions[bot], @Mariefay

**Full changelog**: https://github.com/triggerdotdev/trigger.dev/compare/v4.7.1...v4.7.2

Related Repositories