Skip to content

fix(agent): broadcast tool.result per call as it settles, not after the round #117

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
w0nche0l wants to merge 3 commits into
base: main
Choose a base branch
from
+738 −158
Open

fix(agent): broadcast tool.result per call as it settles, not after the round #117

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3e47eec
fix(agent): broadcast tool.result per call as it settles, not after t…
w0nche0l Aug 25, 2026
75807ec
Merge origin/main into dennisjeong/dev-1067-broadcast-tool-result-per…
LukasParke Oct 21, 2026
cc54771
fix(agent): settle all round calls before rejecting; order defer clai…
LukasParke Oct 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions .changeset/tool-result-per-call-settle.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
---
'@openrouter/agent': patch
---

Broadcast each tool call's `tool.result` / `tool.call_output` (and the legacy `tool_result`) the moment that call settles, instead of holding every broadcast until the whole round's `Promise.allSettled` resolves. With serially executing calls (e.g. `maxConcurrency: 1`), stream consumers now see each call complete live rather than all at once when the round ends. (DEV-1067)

**Behaviour change for stream consumers:** within a round, `tool.call_output`, `tool.result` and legacy `tool_result` events now arrive in **completion order**, not call order. Correlate them by `callId` / `toolCallId` rather than by position. Model-facing `function_call_output` items are still assembled in **call order**, so follow-up request input and prompt-cache behaviour don't change.

If a call's `toModelOutput` throws, the round still waits for every sibling call to settle before it rejects with the first error in call order, so no broadcasts or async registrations land after the run has failed. When two `ctx.defer()` calls in the same round claim the same taskId, the earlier call in call order still wins. To make that deterministic, a deferred call's registration and placeholder broadcast wait for earlier calls in the round to settle.
Loading
Loading