[BUG] time consuming tools are crashing in cloudflare workers

[sapkra]

Describe the bug

Claude says:
VoltAgent's withWaitUntil sets the global ___voltagent_wait_until inside toCloudflareWorker().fetch() but cleans it up in finally as soon as the response object is returned — before streaming/tools complete. The tools then crash because the global waitUntil has been cleared.

The workaround: pass ctx.waitUntil through the context so tools can call setWaitUntil to re-register it.

Steps To Reproduce

use voltagent in a cloudflare worker with an agent that calls a tool that takes a few seconds

Expected behavior

agent doesn't crash after tool-input-available event

Packages

"@voltagent/cli": "0.1.21",
"@voltagent/core": "2.6.14",
"@voltagent/logger": "2.0.2",
"@voltagent/resumable-streams": "2.0.2",
"@voltagent/server-core": "2.1.11",
"@voltagent/server-hono": "2.0.8",
"@voltagent/serverless-hono": "2.0.9",

Additional Context

No response

[ravyg]

We ran into this same issue running VoltAgent on Cloudflare Workers with tools that take a few seconds to complete. Traced it to the finally block in toCloudflareWorker() calling cleanup() before streaming and tool execution finish.

Opened a PR with a fix: #1191

The fix defers cleanup through the platform's own waitUntil() so the global stays alive until all background work settles. Falls back to synchronous cleanup for non-serverless environments. Includes 7 tests and a changeset.