Merge Modelfamily into modelinfo (#8763)

- Merge ModelFamily into ModelInfo
- Remove logic for adding instructions to apply patch
- Add compaction limit and visible context window to `ModelInfo`

Author: aibrahim-oai

codex-rs/app-server-protocol/src/protocol/v2.rs

@@ -1247,6 +1247,7 @@ pub struct ThreadTokenUsageUpdatedNotification {
 pub struct ThreadTokenUsage {
     pub total: TokenUsageBreakdown,
     pub last: TokenUsageBreakdown,
+    // TODO(aibrahim): make this not optional
     #[ts(type = "number | null")]
     pub model_context_window: Option<i64>,
 }

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

@@ -15,7 +15,7 @@ fn preset_to_info(preset: &ModelPreset, priority: i32) -> ModelInfo {
         slug: preset.id.clone(),
         display_name: preset.display_name.clone(),
         description: Some(preset.description.clone()),
-        default_reasoning_level: preset.default_reasoning_effort,
+        default_reasoning_level: Some(preset.default_reasoning_effort),
         supported_reasoning_levels: preset.supported_reasoning_efforts.clone(),
         shell_type: ConfigShellToolType::ShellCommand,
         visibility: if preset.show_in_picker {
@@ -26,14 +26,16 @@ fn preset_to_info(preset: &ModelPreset, priority: i32) -> ModelInfo {
         supported_in_api: true,
         priority,
         upgrade: preset.upgrade.as_ref().map(|u| u.id.clone()),
-        base_instructions: None,
+        base_instructions: "base instructions".to_string(),
         supports_reasoning_summaries: false,
         support_verbosity: false,
         default_verbosity: None,
         apply_patch_tool_type: None,
         truncation_policy: TruncationPolicyConfig::bytes(10_000),
         supports_parallel_tool_calls: false,
-        context_window: None,
+        context_window: Some(272_000),
+        auto_compact_token_limit: None,
+        effective_context_window_percent: 95,
         experimental_supported_tools: Vec::new(),
     }
 }

codex-rs/codex-api/src/endpoint/models.rs

@@ -215,14 +215,14 @@ mod tests {
                     "supported_in_api": true,
                     "priority": 1,
                     "upgrade": null,
-                    "base_instructions": null,
+                    "base_instructions": "base instructions",
                     "supports_reasoning_summaries": false,
                     "support_verbosity": false,
                     "default_verbosity": null,
                     "apply_patch_tool_type": null,
                     "truncation_policy": {"mode": "bytes", "limit": 10_000},
                     "supports_parallel_tool_calls": false,
-                    "context_window": null,
+                    "context_window": 272_000,
                     "experimental_supported_tools": [],
                 }))
                 .unwrap(),

codex-rs/codex-api/tests/models_integration.rs

@@ -56,7 +56,7 @@ async fn models_client_hits_models_endpoint() {
             slug: "gpt-test".to_string(),
             display_name: "gpt-test".to_string(),
             description: Some("desc".to_string()),
-            default_reasoning_level: ReasoningEffort::Medium,
+            default_reasoning_level: Some(ReasoningEffort::Medium),
             supported_reasoning_levels: vec![
                 ReasoningEffortPreset {
                     effort: ReasoningEffort::Low,
@@ -76,14 +76,16 @@ async fn models_client_hits_models_endpoint() {
             supported_in_api: true,
             priority: 1,
             upgrade: None,
-            base_instructions: None,
+            base_instructions: "base instructions".to_string(),
             supports_reasoning_summaries: false,
             support_verbosity: false,
             default_verbosity: None,
             apply_patch_tool_type: None,
             truncation_policy: TruncationPolicyConfig::bytes(10_000),
             supports_parallel_tool_calls: false,
-            context_window: None,
+            context_window: Some(272_000),
+            auto_compact_token_limit: None,
+            effective_context_window_percent: 95,
             experimental_supported_tools: Vec::new(),
         }],
     };

codex-rs/core/prompt_with_apply_patch_instructions.md

@@ -22,6 +22,7 @@ use codex_otel::otel_manager::OtelManager;
 use codex_protocol::ThreadId;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::openai_models::ModelInfo;
 use codex_protocol::openai_models::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::protocol::SessionSource;
 use eventsource_stream::Event;
@@ -49,15 +50,14 @@ use crate::features::FEATURES;
 use crate::flags::CODEX_RS_SSE_FIXTURE;
 use crate::model_provider_info::ModelProviderInfo;
 use crate::model_provider_info::WireApi;
-use crate::models_manager::model_family::ModelFamily;
 use crate::tools::spec::create_tools_json_for_chat_completions_api;
 use crate::tools::spec::create_tools_json_for_responses_api;
 
 #[derive(Debug, Clone)]
 pub struct ModelClient {
     config: Arc<Config>,
     auth_manager: Option<Arc<AuthManager>>,
-    model_family: ModelFamily,
+    model_info: ModelInfo,
     otel_manager: OtelManager,
     provider: ModelProviderInfo,
     conversation_id: ThreadId,
@@ -71,7 +71,7 @@ impl ModelClient {
     pub fn new(
         config: Arc<Config>,
         auth_manager: Option<Arc<AuthManager>>,
-        model_family: ModelFamily,
+        model_info: ModelInfo,
         otel_manager: OtelManager,
         provider: ModelProviderInfo,
         effort: Option<ReasoningEffortConfig>,
@@ -82,7 +82,7 @@ impl ModelClient {
         Self {
             config,
             auth_manager,
-            model_family,
+            model_info,
             otel_manager,
             provider,
             conversation_id,
@@ -93,11 +93,11 @@ impl ModelClient {
     }
 
     pub fn get_model_context_window(&self) -> Option<i64> {
-        let model_family = self.get_model_family();
-        let effective_context_window_percent = model_family.effective_context_window_percent;
-        model_family
-            .context_window
-            .map(|w| w.saturating_mul(effective_context_window_percent) / 100)
+        let model_info = self.get_model_info();
+        let effective_context_window_percent = model_info.effective_context_window_percent;
+        model_info.context_window.map(|context_window| {
+            context_window.saturating_mul(effective_context_window_percent) / 100
+        })
     }
 
     pub fn config(&self) -> Arc<Config> {
@@ -146,8 +146,8 @@ impl ModelClient {
         }
 
         let auth_manager = self.auth_manager.clone();
-        let model_family = self.get_model_family();
-        let instructions = prompt.get_full_instructions(&model_family).into_owned();
+        let model_info = self.get_model_info();
+        let instructions = prompt.get_full_instructions(&model_info).into_owned();
         let tools_json = create_tools_json_for_chat_completions_api(&prompt.tools)?;
         let api_prompt = build_api_prompt(prompt, instructions, tools_json);
         let conversation_id = self.conversation_id.to_string();
@@ -200,13 +200,14 @@ impl ModelClient {
         }
 
         let auth_manager = self.auth_manager.clone();
-        let model_family = self.get_model_family();
-        let instructions = prompt.get_full_instructions(&model_family).into_owned();
+        let model_info = self.get_model_info();
+        let instructions = prompt.get_full_instructions(&model_info).into_owned();
         let tools_json: Vec<Value> = create_tools_json_for_responses_api(&prompt.tools)?;
 
-        let reasoning = if model_family.supports_reasoning_summaries {
+        let default_reasoning_effort = model_info.default_reasoning_level;
+        let reasoning = if model_info.supports_reasoning_summaries {
             Some(Reasoning {
-                effort: self.effort.or(model_family.default_reasoning_effort),
+                effort: self.effort.or(default_reasoning_effort),
                 summary: if self.summary == ReasoningSummaryConfig::None {
                     None
                 } else {
@@ -223,15 +224,13 @@ impl ModelClient {
             vec![]
         };
 
-        let verbosity = if model_family.support_verbosity {
-            self.config
-                .model_verbosity
-                .or(model_family.default_verbosity)
+        let verbosity = if model_info.support_verbosity {
+            self.config.model_verbosity.or(model_info.default_verbosity)
         } else {
             if self.config.model_verbosity.is_some() {
                 warn!(
                     "model_verbosity is set but ignored as the model does not support verbosity: {}",
-                    model_family.family
+                    model_info.slug
                 );
             }
             None
@@ -298,12 +297,11 @@ impl ModelClient {
 
     /// Returns the currently configured model slug.
     pub fn get_model(&self) -> String {
-        self.get_model_family().get_model_slug().to_string()
+        self.model_info.slug.clone()
     }
 
-    /// Returns the currently configured model family.
-    pub fn get_model_family(&self) -> ModelFamily {
-        self.model_family.clone()
+    pub fn get_model_info(&self) -> ModelInfo {
+        self.model_info.clone()
     }
 
     /// Returns the current reasoning effort setting.
@@ -340,7 +338,7 @@ impl ModelClient {
             .with_telemetry(Some(request_telemetry));
 
         let instructions = prompt
-            .get_full_instructions(&self.get_model_family())
+            .get_full_instructions(&self.get_model_info())
             .into_owned();
         let payload = ApiCompactionInput {
             model: &self.get_model(),

codex-rs/core/src/client.rs

@@ -1,15 +1,13 @@
 use crate::client_common::tools::ToolSpec;
 use crate::error::Result;
-use crate::models_manager::model_family::ModelFamily;
 pub use codex_api::common::ResponseEvent;
-use codex_apply_patch::APPLY_PATCH_TOOL_INSTRUCTIONS;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::openai_models::ModelInfo;
 use futures::Stream;
 use serde::Deserialize;
 use serde_json::Value;
 use std::borrow::Cow;
 use std::collections::HashSet;
-use std::ops::Deref;
 use std::pin::Pin;
 use std::task::Context;
 use std::task::Poll;
@@ -44,28 +42,12 @@ pub struct Prompt {
 }
 
 impl Prompt {
-    pub(crate) fn get_full_instructions<'a>(&'a self, model: &'a ModelFamily) -> Cow<'a, str> {
-        let base = self
-            .base_instructions_override
-            .as_deref()
-            .unwrap_or(model.base_instructions.deref());
-        // When there are no custom instructions, add apply_patch_tool_instructions if:
-        // - the model needs special instructions (4.1)
-        // AND
-        // - there is no apply_patch tool present
-        let is_apply_patch_tool_present = self.tools.iter().any(|tool| match tool {
-            ToolSpec::Function(f) => f.name == "apply_patch",
-            ToolSpec::Freeform(f) => f.name == "apply_patch",
-            _ => false,
-        });
-        if self.base_instructions_override.is_none()
-            && model.needs_special_apply_patch_instructions
-            && !is_apply_patch_tool_present
-        {
-            Cow::Owned(format!("{base}\n{APPLY_PATCH_TOOL_INSTRUCTIONS}"))
-        } else {
-            Cow::Borrowed(base)
-        }
+    pub(crate) fn get_full_instructions<'a>(&'a self, model: &'a ModelInfo) -> Cow<'a, str> {
+        Cow::Borrowed(
+            self.base_instructions_override
+                .as_deref()
+                .unwrap_or(model.base_instructions.as_str()),
+        )
     }
 
     pub(crate) fn get_formatted_input(&self) -> Vec<ResponseItem> {
@@ -277,6 +259,8 @@ mod tests {
         let prompt = Prompt {
             ..Default::default()
         };
+        let prompt_with_apply_patch_instructions =
+            include_str!("../prompt_with_apply_patch_instructions.md");
         let test_cases = vec![
             InstructionsTestCase {
                 slug: "gpt-3.5",
@@ -317,19 +301,16 @@ mod tests {
         ];
         for test_case in test_cases {
             let config = test_config();
-            let model_family =
-                ModelsManager::construct_model_family_offline(test_case.slug, &config);
-            let expected = if test_case.expects_apply_patch_instructions {
-                format!(
-                    "{}\n{}",
-                    model_family.clone().base_instructions,
-                    APPLY_PATCH_TOOL_INSTRUCTIONS
-                )
-            } else {
-                model_family.clone().base_instructions
-            };
-
-            let full = prompt.get_full_instructions(&model_family);
+            let model_info = ModelsManager::construct_model_info_offline(test_case.slug, &config);
+            if test_case.expects_apply_patch_instructions {
+                assert_eq!(
+                    model_info.base_instructions.as_str(),
+                    prompt_with_apply_patch_instructions
+                );
+            }
+
+            let expected = model_info.base_instructions.as_str();
+            let full = prompt.get_full_instructions(&model_info);
             assert_eq!(full, expected);
         }
     }