Add item streaming events #5546

pakrym-oai · 2025-10-23T00:20:26Z

Adds AgentMessageContentDelta, ReasoningContentDelta, ReasoningRawContentDelta item streaming events while maintaining compatibility for old events.

- Introduce AgentMessageContentDeltaEvent, ReasoningContentDeltaEvent, and ReasoningRawContentDeltaEvent with thread_id, turn_id, and item_id, emitted during streaming, alongside legacy events for compatibility. - Refactor output delta handling to associate deltas with active items and emit item-specific delta events. - Move legacy event emission via new HasLegacyEvent trait for simpler event compatibility. - Update parse_agent_message to use

# Conflicts: # codex-rs/core/src/codex.rs

…egacy event for ItemStartedEvent - Move `error_or_panic` from conversation_history.rs to util.rs and update all usage sites accordingly. - Use `error_or_panic` in codex.rs to handle error conditions on various response events instead of directly logging with `error!`. - Implement `HasLegacyEvent` for `ItemStartedEvent` in protocol.rs to emit a WebSearchBegin legacy event

…xt message

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2025-10-24T02:50:25Z

codex-rs/core/src/chat_completions.rs

+        let item = ResponseItem::Message {
+            id: None,
+            role: "assistant".to_string(),
+            content: vec![],
+        };


Emit stable IDs for streamed assistant items

When an assistant delta arrives in chat completions, append_assistant_text builds a ResponseItem::Message with id: None. Downstream, parse_turn_item assigns a fresh UUID whenever an item lacks an ID, so ItemStarted, AgentMessageContentDelta, and ItemCompleted events for the same streamed message use different generated IDs. Clients can’t correlate deltas with their corresponding completion and tests only pass because the mock SSE events already provide IDs. Assign a concrete ID when the item is first constructed and reuse it for all subsequent events so the identifier stays consistent.

Useful? React with 👍 / 👎.

seems potentially legit feedback?

ok so the feedback is legit but it always worked this way.

@pakrym-oai how big of a deal is this? either way, addressing this seems out of scope for this PR

As long as we don't emit Item events with different item id (which I don't think we do based on active.id()) we should be good.

owenlin0

overall lgtm! just a comment about the web search call

owenlin0 · 2025-10-24T15:32:31Z

codex-rs/protocol/src/protocol.rs

+
+#[derive(Debug, Clone, Deserialize, Serialize, TS, JsonSchema)]
+pub struct AgentMessageContentDeltaEvent {
+    pub thread_id: String,


decided to add these to all the events? nice!

owenlin0 · 2025-10-24T15:40:08Z

codex-rs/core/src/chat_completions.rs

+        let item = ResponseItem::Message {
+            id: None,
+            role: "assistant".to_string(),
+            content: vec![],
+        };


seems potentially legit feedback?

owenlin0 · 2025-10-24T15:48:26Z

codex-rs/core/src/client.rs

-                            .and_then(|v| v.as_str())
-                            .unwrap_or("")
-                            .to_string();
-                        let ev = ResponseEvent::WebSearchCallBegin { call_id };


I noticed we delete the WebSearchCall event in various places - what's the plan for that one?

Are we planning to support some translation of https://platform.openai.com/docs/api-reference/responses-streaming/response/web_search_call or just dropping it entirely?

My guess is Responses API does something like this for a web search call:

response.output_item.added response.web_search_call.in_progress response.web_search_call.searching response.web_search_call.completed response.output_item.done

I think we don't actually need to handle web_search_call events specifically and we can emit EventMsg::WebSearchBegin and EventMsg::WebSearchEnd based on response.output_item.added response.output_item.done

Here's a trace of an actual request:

�[2m2025-10-24T16:10:55.257490Z�[0m �[35mTRACE�[0m SSE event: {"type":"response.output_item.added","sequence_number":203,"output_index":1,"item":{"id":"ws_0bb4a7dd4caece510168fba50f3a788190a16bcfee2fcef85a","type":"web_search_call","status":"in_progress"}} �[2m2025-10-24T16:10:55.267924Z�[0m �[35mTRACE�[0m SSE event: {"type":"response.web_search_call.in_progress","sequence_number":204,"output_index":1,"item_id":"ws_0bb4a7dd4caece510168fba50f3a788190a16bcfee2fcef85a"} �[2m2025-10-24T16:10:56.830551Z�[0m �[35mTRACE�[0m SSE event: {"type":"response.web_search_call.searching","sequence_number":205,"output_index":1,"item_id":"ws_0bb4a7dd4caece510168fba50f3a788190a16bcfee2fcef85a"} �[2m2025-10-24T16:11:02.043478Z�[0m �[35mTRACE�[0m SSE event: {"type":"response.web_search_call.completed","sequence_number":206,"output_index":1,"item_id":"ws_0bb4a7dd4caece510168fba50f3a788190a16bcfee2fcef85a"} �[2m2025-10-24T16:11:02.046008Z�[0m �[35mTRACE�[0m SSE event: {"type":"response.output_item.done","sequence_number":207,"output_index":1,"item":{"id":"ws_0bb4a7dd4caece510168fba50f3a788190a16bcfee2fcef85a","type":"web_search_call","status":"completed","action":{"type":"search","query":"top news headlines October 24, 2025 Reuters"}}}

So Codex would do this client-side? i.e. Codex would inspect the output_item when it's added, and if it's a web search call, emit the EventMsg::WebSearchBegin event? I think that'd work but why not just translate the server events? Seems cleaner that way

So Codex would do this client-side?
We already do this. see https://github.com/openai/codex/pull/5546/files#diff-6d2d1446867cf2e611597ebbf293689be55ad7cd32c1081cdab6de2358dccb91R539

If end goal is to have codex level item_starter/item_completed event and we can do it for all item types consistently by listening to item.added/item.completed from Responses API, why not do it? The end result is the same and the logic is simpler for us. We use the same code path for WebSearch, Agent Messages and Reasoning.

ok sounds good. do we plan to keep exposing WebSearchCall events in the stable app-server API too?

No, only ItemStarted with a WebSearch item type

oh i see, then it makes a lot of sense to do it client-side. ty for the explanation! 🚢

owenlin0 · 2025-10-29T22:24:54Z

codex-rs/core/src/tasks/review.rs

+            EventMsg::ItemCompleted(ItemCompletedEvent {
+                item: TurnItem::AgentMessage(_),
+                ..
+            }) => {}


@aibrahim-oai @pakrym-oai is this fix legit? trying to make sense of the review sub-codex delegation flow. codex helped me generate this fix, otherwise review_does_not_emit_agent_message_on_structured_output fails

yeah, reasonable

pakrym-oai added 5 commits October 22, 2025 17:20

Add item streaming events

e3e90b8

Merge branch 'main' into pakrym/add-item-streaming-events

7bf6175

# Conflicts: # codex-rs/core/src/codex.rs

tests/client: add SSE stream test case with empty assistant output te…

d846b4f

…xt message

pakrym-oai marked this pull request as ready for review October 24, 2025 02:47

chatgpt-codex-connector bot reviewed Oct 24, 2025

View reviewed changes

owenlin0 approved these changes Oct 24, 2025

View reviewed changes

Merge branch 'main' into pakrym/add-item-streaming-events

8658848

owenlin0 force-pushed the pakrym/add-item-streaming-events branch 2 times, most recently from a08f260 to 8658848 Compare October 28, 2025 23:55

owenlin0 added 4 commits October 29, 2025 14:12

Merge branch 'main' into pakrym/add-item-streaming-events

ed35ff5

fix

6449a99

Merge branch 'main' into pakrym/add-item-streaming-events

12f459f

fix

0aa5382

owenlin0 enabled auto-merge (squash) October 29, 2025 21:54

fix?

1ffabd5

owenlin0 reviewed Oct 29, 2025

View reviewed changes

owenlin0 merged commit 3429e82 into main Oct 29, 2025
25 checks passed

owenlin0 deleted the pakrym/add-item-streaming-events branch October 29, 2025 22:33

github-actions bot locked and limited conversation to collaborators Oct 29, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add item streaming events #5546

Add item streaming events #5546

Uh oh!

pakrym-oai commented Oct 23, 2025 •

edited

Loading

chatgpt-codex-connector bot left a comment

chatgpt-codex-connector bot Oct 24, 2025

owenlin0 Oct 24, 2025

owenlin0 Oct 28, 2025

pakrym-oai Oct 29, 2025

owenlin0 left a comment

owenlin0 Oct 24, 2025

owenlin0 Oct 24, 2025

owenlin0 Oct 24, 2025

pakrym-oai Oct 24, 2025

owenlin0 Oct 24, 2025

pakrym-oai Oct 24, 2025

owenlin0 Oct 24, 2025

pakrym-oai Oct 24, 2025

owenlin0 Oct 24, 2025

owenlin0 Oct 29, 2025

pakrym-oai Oct 29, 2025

Uh oh!

Labels

3 participants

Add item streaming events #5546

Add item streaming events #5546

Uh oh!

Conversation

pakrym-oai commented Oct 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

chatgpt-codex-connector bot Oct 24, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

owenlin0 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Uh oh!

Labels

3 participants

pakrym-oai commented Oct 23, 2025 •

edited

Loading