make `model` optional in config #7769

aibrahim-oai · 2025-12-09T08:20:26Z

Make Config.model optional and centralize default-selection logic in ModelsManager, including a default_model helper (with codex-auto-balanced when available) so sessions now carry an explicit chosen model separate from the base config.
Resolve model once in core and tui from config. Then store the state of it on other structs.
Move refreshing models to be before resolving the default model

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".

codex-rs/core/src/codex.rs

aibrahim-oai · 2025-12-09T08:32:52Z

@codex review this

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".

codex-rs/exec/src/lib.rs

codex-rs/tui/src/app.rs

codex-rs/core/src/config/mod.rs

codex-rs/core/src/openai_models/models_manager.rs

codex-rs/core/src/sandboxing/assessment.rs

codex-rs/core/src/client.rs

codex-rs/core/src/codex.rs

pakrym-oai · 2025-12-09T16:07:44Z

codex-rs/core/src/codex.rs

    fn build_per_turn_config(session_configuration: &SessionConfiguration) -> Config {
        let config = session_configuration.original_config_do_not_use.clone();
        let mut per_turn_config = (*config).clone();
-        per_turn_config.model = session_configuration.model.clone();


are we confident nobody is reading this?

Supposedly yeah. I replaced it with turn context

pakrym-oai · 2025-12-09T16:08:26Z

codex-rs/core/src/codex.rs


        let model_family = models_manager
-            .construct_model_family(&config.model, &config)
+            .construct_model_family(&session_configuration.model, &config)


I'd love if everything model-related happend through model_family and we didn't flow raw string model anywhere.

I'll do that

codex-rs/core/tests/suite/client.rs

codex-rs/core/tests/suite/prompt_caching.rs

pakrym-oai · 2025-12-09T16:17:34Z

codex-rs/core/tests/suite/review.rs


-    // Assert the request body model equals the configured review model
+    // Assert the review request currently uses the primary chat model.
+    // TODO: switch back to asserting the custom review model once the runtime plumbs it through.


Is there a way we don't merge broken default review model?

codex-rs/lmstudio/src/lib.rs

codex-rs/ollama/src/lib.rs

codex-rs/tui/src/chatwidget/tests.rs

pakrym-oai · 2025-12-09T16:36:59Z

codex-rs/tui/src/status/tests.rs

        Some(&rate_display),
        None,
        captured_at,
+        config_model(&config),


very roundabout way to pass the model into this method. Does it have to be set on config for some other reason?

pakrym-oai

I think having a separate default_model method on models manager is a footgun.

Models manager should have 3 methods:
async list_models
async get_model(requested_model: Optional<...>)

both of which should async block on manager waiting for models to get loaded.

we can use async list_models as a way to trigger warm up of model manager on start.

I'm also a bit worried about all the callsites that read model override fields from config directly and us having no systematic way of getting rid of them.

aibrahim-oai · 2025-12-10T02:51:07Z

@codex review this

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".

codex-rs/core/src/openai_models/models_manager.rs

aibrahim-oai · 2025-12-10T03:18:22Z

@codex review this. something different than Remote model refresh ignores configured provider

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 · 2025-12-10T03:29:31Z

codex-rs/core/src/codex.rs

        .client
        .get_otel_event_manager()
        .with_model(
-            per_turn_config.model.as_str(),
+            parent_turn_context.client.get_model().as_str(),
            review_model_family.slug.as_str(),


Tag review telemetry with the review model

Review threads now create the OTEL event manager with parent_turn_context.client.get_model() even though the review task actually runs against review_model_family (the dedicated review model). This sets the telemetry model attribute to the conversation’s main model while the slug is the review model, so /review traffic will be misattributed in monitoring and usage reporting. Use the review model slug for both parameters to keep telemetry consistent with the model that is actually invoked.

Useful? React with 👍 / 👎.

pakrym-oai · 2025-12-10T04:09:47Z

codex-rs/app-server/tests/common/models_cache.rs

+use std::path::Path;
+
+/// Convert a ModelPreset to ModelInfo for cache storage.
+fn preset_to_info(preset: &ModelPreset, priority: i32) -> ModelInfo {


surprised to find this file in app-server tests

pakrym-oai · 2025-12-10T04:10:30Z

codex-rs/app-server/tests/suite/v2/model_list.rs

 #[tokio::test]
 async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
    let codex_home = TempDir::new()?;
+    write_models_cache(codex_home.path())?;


do we enable remote models in this test?

Maybe we should have some default mock for model list call? Pretty strange to have to workaround this via writing a cache in an unrelated test.

pakrym-oai · 2025-12-10T04:13:31Z

codex-rs/tui/src/history_cell.rs

-            ModelsManager::construct_model_family_offline(&config.model, &config)
+            ModelsManager::construct_model_family_offline(&default_model, &config)
                .reasoning_summary_format;
        let cell = new_reasoning_summary_block(


not for this PR but we can just hardcode reasoning format here.

pakrym-oai · 2025-12-10T04:15:38Z

codex-rs/tui/src/chatwidget.rs

        } = common;
+        let model_slug = model_family.get_model_slug().to_string();
+        let mut config = config;
+        config.model = Some(model_slug.clone());


the fact that we set the model here and it looks like someplace cares about but never change it again makes me nervous but that's just a feeling.

Who cares about this value?

pakrym-oai · 2025-12-10T04:17:02Z

codex-rs/tui/src/app_backtrack.rs

    ) {
        let conv = new_conv.conversation;
        let session_configured = new_conv.session_configured;
+        let model_family = self.chat_widget.get_model_family();


roundabout way to pass model family in but fine I guess.

pakrym-oai · 2025-12-10T04:19:15Z

codex-rs/tui/src/app.rs

                    skills: self.skills.clone(),
                    is_first_run: false,
-                    model_family,
+                    model_family: model_family.clone(),


Will this bite us? This feels like ChatWidgetInit needs model family way too early..

pakrym-oai · 2025-12-10T04:21:49Z

codex-rs/tui/src/app.rs

                    migration_config: migration_config_key.to_string(),
                });
-                config.model = target_model.to_string();
+                config.model = Some(target_model.clone());


At this point we maybe should just return the new model but that's for the future.

pakrym-oai · 2025-12-10T04:22:31Z

codex-rs/exec/src/lib.rs

    let conversation_manager = ConversationManager::new(auth_manager.clone(), SessionSource::Exec);
+    let default_model = conversation_manager
+        .get_models_manager()
+        .get_model(config.model.clone().as_deref(), &config)


nit: for later, can this take &Option so we avoid all these clones?

pakrym-oai · 2025-12-10T04:23:26Z

codex-rs/core/tests/chat_completions_payload.rs

+    let all_requests = server
+        .received_requests()
+        .await
+        .unwrap_or_else(|| panic!("received requests"));


are you trying to hide .expect?

pakrym-oai · 2025-12-10T04:24:35Z

codex-rs/core/tests/suite/rmcp_client.rs

    // Chat Completions assertion: the second POST should include a tool role message
    // with an array `content` containing an item with the expected data URL.
-    let requests = server.received_requests().await.expect("requests captured");
+    let all_requests = server.received_requests().await.expect("requests captured");


Is this because we mock models endpoint? Why are remote models enabled in these tests?

pakrym-oai · 2025-12-10T04:25:24Z

codex-rs/core/tests/suite/review.rs

    // Assert the request body model equals the configured review model
-    let request = &server.received_requests().await.unwrap()[0];
+    let requests = get_responses_requests(&server).await;
+    let request = requests


Same here, can we pick a smaller subset of tests to enable remote_models for?

aibrahim-oai · 2025-12-10T05:49:30Z

@codex review this. something different than Remote model refresh ignores configured provider

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 · 2025-12-10T05:57:02Z

codex-rs/core/src/codex.rs

        .with_model(
-            per_turn_config.model.as_str(),
+            parent_turn_context.client.get_model().as_str(),
            review_model_family.slug.as_str(),


Use review model when tagging OTEL metadata

Review threads are instrumented under the main session model even though the client uses review_model_family. In spawn_review_thread the OTEL event manager is built with parent_turn_context.client.get_model() instead of the review model, so whenever the review model differs from the conversation model the telemetry/logs will attribute review activity to the wrong model even though the actual requests use the review model.

Useful? React with 👍 / 👎.

aibrahim-oai added 4 commits December 8, 2025 22:29

config_model

1dc2e61

progress

1f08695

test

b18fef0

test

9aaf6a0

chatgpt-codex-connector bot reviewed Dec 9, 2025

View reviewed changes

codex-rs/core/src/codex.rs Outdated Show resolved Hide resolved

chatgpt-codex-connector bot reviewed Dec 9, 2025

View reviewed changes

codex-rs/exec/src/lib.rs Outdated Show resolved Hide resolved

codex-rs/tui/src/app.rs Outdated Show resolved Hide resolved