tkmind/tkmind_go
1
0
Fork 0
Code Issues Pull Requests Actions 4 Packages Projects Releases Wiki Activity

Surface resolved Databricks model metadata (#9206)

Browse Source 
Signed-off-by: jh-block <jhugo@block.xyz>
This commit is contained in:
jh-block
2026-05-20 12:50:44 +02:00
committed by GitHub
parent 98a54e9ec6
commit 8f9270e295
19 changed files with 369 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/goose-server/src/openapi.rs
+5 -3
View File
@@ -20,9 +20,10 @@ use goose::config::declarative_providers::{
DeclarativeProviderConfig, EnvVarConfig, LoadedProvider, ProviderEngine,
};
use goose::conversation::message::{
ActionRequired, ActionRequiredData, FrontendToolRequest, Message, MessageContent,
MessageMetadata, RedactedThinkingContent, SystemNotificationContent, SystemNotificationType,
ThinkingContent, TokenState, ToolConfirmationRequest, ToolRequest, ToolResponse,
ActionRequired, ActionRequiredData, FrontendToolRequest, InferenceMetadata, Message,
MessageContent, MessageMetadata, RedactedThinkingContent, SystemNotificationContent,
SystemNotificationType, ThinkingContent, TokenState, ToolConfirmationRequest, ToolRequest,
ToolResponse,
};
use crate::routes::recipe_utils::RecipeManifest;
@@ -528,6 +529,7 @@ derive_utoipa!(IconTheme as IconThemeSchema);
Message,
MessageContent,
MessageMetadata,
InferenceMetadata,
TokenState,
ContentSchema,
EmbeddedResourceSchema,
crates/goose/src/agents/agent.rs
+36 -2
View File
@@ -34,8 +34,8 @@ use crate::context_mgmt::{
check_if_compaction_needed, compact_messages, DEFAULT_COMPACTION_THRESHOLD,
};
use crate::conversation::message::{
ActionRequiredData, Message, MessageContent, ProviderMetadata, SystemNotificationType,
ToolRequest,
ActionRequiredData, InferenceMetadata, Message, MessageContent, ProviderMetadata,
SystemNotificationType, ToolRequest,
};
use crate::conversation::{debug_conversation_fix, fix_conversation, Conversation};
use crate::mcp_utils::ToolResult;
@@ -1586,9 +1586,22 @@ impl Agent {
self.reset_retry_attempts().await;
let provider = self.provider().await?;
let provider_name = provider.get_name().to_string();
let requested_model = provider.get_model_config().model_name;
let inference = provider
.fetch_model_info(&requested_model)
.await
.ok()
.and_then(|model_info| model_info.resolved_model)
.map(|resolved_model| InferenceMetadata {
provider: provider_name,
requested_model,
resolved_model: Some(resolved_model),
});
let session_manager = self.config.session_manager.clone();
let session_id = session_config.id.clone();
if !self.config.disable_session_naming {
let provider = provider.clone();
let manager_for_spawn = session_manager.clone();
let session_name_update_tx = self.config.session_name_update_tx.clone();
tokio::spawn(async move {
@@ -1726,6 +1739,17 @@ impl Agent {
)
.await;
let filtered_response = if let Some(inference) = inference.as_ref() {
filtered_response.with_inference(inference.clone())
} else {
filtered_response
};
let response = if let Some(inference) = inference.as_ref() {
response.with_inference(inference.clone())
} else {
response
};
surfaced_thinking_in_turn |= filtered_response.content.iter().any(
|content| {
matches!(
@@ -2234,6 +2258,16 @@ impl Agent {
}
}
let messages_to_add = if let Some(ref inference) = inference {
Conversation::new_unvalidated(
messages_to_add
.into_iter()
.map(|message| message.with_inference_if_assistant(inference.clone())),
)
} else {
messages_to_add
};
for msg in &messages_to_add {
session_manager.add_message(&session_config.id, msg).await?;
}
crates/goose/src/context_mgmt/mod.rs
+1 -1
View File
@@ -143,7 +143,7 @@ pub async fn compact_messages(
// This is the most recent message and we're preserving it by adding a fresh copy
MessageMetadata::invisible()
} else {
msg.metadata.with_agent_invisible()
msg.metadata.clone().with_agent_invisible()
};
let updated_msg = msg.clone().with_metadata(updated_metadata);
final_messages.push(updated_msg);
crates/goose/src/conversation/message.rs
+35 -2
View File
@@ -641,14 +641,25 @@ impl From<PromptMessage> for Message {
}
}
#[derive(ToSchema, Clone, Copy, PartialEq, Serialize, Deserialize, Debug)]
/// Metadata for message visibility
#[derive(ToSchema, Clone, PartialEq, Serialize, Deserialize, Debug)]
#[serde(rename_all = "camelCase")]
pub struct InferenceMetadata {
pub provider: String,
pub requested_model: String,
#[serde(skip_serializing_if = "Option::is_none")]
pub resolved_model: Option<String>,
}
#[derive(ToSchema, Clone, PartialEq, Serialize, Deserialize, Debug)]
/// Metadata for message visibility and model inference details
#[serde(rename_all = "camelCase")]
pub struct MessageMetadata {
/// Whether the message should be visible to the user in the UI
pub user_visible: bool,
/// Whether the message should be included in the agent's context window
pub agent_visible: bool,
#[serde(skip_serializing_if = "Option::is_none")]
pub inference: Option<InferenceMetadata>,
}
impl Default for MessageMetadata {
@@ -656,6 +667,7 @@ impl Default for MessageMetadata {
MessageMetadata {
user_visible: true,
agent_visible: true,
inference: None,
}
}
}
@@ -666,6 +678,7 @@ impl MessageMetadata {
MessageMetadata {
user_visible: false,
agent_visible: true,
..Default::default()
}
}
@@ -674,6 +687,7 @@ impl MessageMetadata {
MessageMetadata {
user_visible: true,
agent_visible: false,
..Default::default()
}
}
@@ -682,6 +696,7 @@ impl MessageMetadata {
MessageMetadata {
user_visible: false,
agent_visible: false,
..Default::default()
}
}
@@ -716,6 +731,11 @@ impl MessageMetadata {
..self
}
}
pub fn with_inference(mut self, inference: InferenceMetadata) -> Self {
self.inference = Some(inference);
self
}
}
#[derive(ToSchema, Clone, PartialEq, Serialize, Deserialize, Debug)]
@@ -996,6 +1016,19 @@ impl Message {
self
}
pub fn with_inference(mut self, inference: InferenceMetadata) -> Self {
self.metadata = self.metadata.with_inference(inference);
self
}
pub fn with_inference_if_assistant(self, inference: InferenceMetadata) -> Self {
if self.role == Role::Assistant && self.metadata.inference.is_none() {
self.with_inference(inference)
} else {
self
}
}
pub fn user_only(mut self) -> Self {
self.metadata.user_visible = true;
self.metadata.agent_visible = false;
crates/goose/src/conversation/mod.rs
+15
View File
@@ -43,11 +43,26 @@ impl Conversation {
}
pub fn push(&mut self, message: Message) {
if message.content.is_empty() && message.metadata.inference.is_some() {
if let Some(existing) = self
.0
.iter_mut()
.rev()
.find(|m| m.role == message.role && m.is_user_visible())
{
existing.metadata.inference = message.metadata.inference;
}
return;
}
if let Some(last) = self
.0
.last_mut()
.filter(|m| m.id.is_some() && m.id == message.id)
{
if message.metadata.inference.is_some() {
last.metadata.inference = message.metadata.inference.clone();
}
match (last.content.last_mut(), message.content.last()) {
(Some(MessageContent::Text(ref mut last)), Some(MessageContent::Text(new)))
if message.content.len() == 1 =>
crates/goose/src/providers/base.rs
+9
View File
@@ -385,6 +385,9 @@ pub static MSG_COUNT_FOR_SESSION_NAME_GENERATION: usize = 3;
pub struct ModelInfo {
/// The name of the model
pub name: String,
/// The underlying model resolved from provider metadata, when the configured model is an alias or endpoint.
#[serde(default, skip_serializing_if = "Option::is_none")]
pub resolved_model: Option<String>,
/// The maximum context length this model supports
pub context_limit: usize,
/// Cost per token for input in USD (optional)
@@ -405,6 +408,7 @@ impl ModelInfo {
pub fn new(name: impl Into<String>, context_limit: usize) -> Self {
Self {
name: name.into(),
resolved_model: None,
context_limit,
input_token_cost: None,
output_token_cost: None,
@@ -423,6 +427,7 @@ impl ModelInfo {
) -> Self {
Self {
name: name.into(),
resolved_model: None,
context_limit,
input_token_cost: Some(input_cost),
output_token_cost: Some(output_cost),
@@ -448,6 +453,7 @@ fn model_info_for_provider_model(provider_name: &str, model_name: &str) -> Model
ModelInfo {
name: model_name.to_string(),
resolved_model: None,
context_limit: ModelConfig::new_or_fail(model_name)
.with_canonical_limits(provider_name)
.context_limit(),
@@ -1778,6 +1784,7 @@ mod tests {
// Test direct ModelInfo creation
let info = ModelInfo {
name: "test-model".to_string(),
resolved_model: None,
context_limit: 1000,
input_token_cost: None,
output_token_cost: None,
@@ -1790,6 +1797,7 @@ mod tests {
// Test equality
let info2 = ModelInfo {
name: "test-model".to_string(),
resolved_model: None,
context_limit: 1000,
input_token_cost: None,
output_token_cost: None,
@@ -1802,6 +1810,7 @@ mod tests {
// Test inequality
let info3 = ModelInfo {
name: "test-model".to_string(),
resolved_model: None,
context_limit: 2000,
input_token_cost: None,
output_token_cost: None,
crates/goose/src/providers/databricks.rs
+9
View File
@@ -539,6 +539,7 @@ impl DatabricksProvider {
ModelInfo {
name: info.name,
resolved_model: info.upstream_model_name,
context_limit,
input_token_cost: None,
output_token_cost: None,
@@ -999,6 +1000,14 @@ mod tests {
assert_eq!(info.name, "goose");
assert_eq!(info.upstream_model_name.as_deref(), Some("claude-opus-4.6"));
assert_eq!(info.reasoning, Some(true));
let model_info = DatabricksProvider::model_info_from_endpoint(info);
assert_eq!(model_info.name, "goose");
assert_eq!(
model_info.resolved_model.as_deref(),
Some("claude-opus-4.6")
);
assert!(model_info.reasoning);
}
#[test]
crates/goose/src/providers/formats/openai.rs
+25 -7
View File
@@ -734,15 +734,22 @@ pub fn get_usage(usage: &Value) -> Usage {
.with_cache_tokens(cache_read_input_tokens, cache_write_input_tokens)
}
fn extract_usage_with_output_tokens(chunk: &StreamingChunk) -> Option<ProviderUsage> {
fn extract_usage_with_output_tokens(
chunk: &StreamingChunk,
fallback_model: Option<&str>,
) -> Option<ProviderUsage> {
chunk
.usage
.as_ref()
.and_then(|u| {
chunk.model.as_ref().map(|model| ProviderUsage {
usage: get_usage(u),
model: model.clone(),
})
chunk
.model
.as_deref()
.or(fallback_model)
.map(|model| ProviderUsage {
usage: get_usage(u),
model: model.to_string(),
})
})
.filter(|u| u.usage.output_tokens.is_some())
}
@@ -901,6 +908,7 @@ where
// reasoning will arrive. Emitting it immediately and then receiving
// reasoning_content in a later chunk would produce duplicated reasoning.
let mut pending_inline_thinking = String::new();
let mut last_seen_model: Option<String> = None;
'outer: while let Some(response) = stream.next().await {
let response_str = response?;
@@ -917,6 +925,9 @@ where
let chunk: StreamingChunk = parse_streaming_chunk(
line.ok_or_else(|| anyhow!("unexpected stream format"))?
)?;
if let Some(model) = &chunk.model {
last_seen_model = Some(model.clone());
}
if !chunk.choices.is_empty() {
if let Some(details) = &chunk.choices[0].delta.reasoning_details {
@@ -931,7 +942,7 @@ where
}
}
let mut usage = extract_usage_with_output_tokens(&chunk);
let mut usage = extract_usage_with_output_tokens(&chunk, last_seen_model.as_deref());
if chunk.choices.is_empty() {
yield (None, usage)
@@ -959,8 +970,11 @@ where
}
let tool_chunk: StreamingChunk = parse_streaming_chunk(line)?;
if let Some(model) = &tool_chunk.model {
last_seen_model = Some(model.clone());
}
if let Some(chunk_usage) = extract_usage_with_output_tokens(&tool_chunk) {
if let Some(chunk_usage) = extract_usage_with_output_tokens(&tool_chunk, last_seen_model.as_deref()) {
usage = Some(chunk_usage);
}
@@ -2528,6 +2542,10 @@ data: [DONE]
assert_eq!(result.tool_calls.len(), 1, "Expected 1 tool call");
assert_eq!(result.tool_calls[0], "developer__shell");
assert_usage_yielded_once(&result, 8320, 172, 8492);
assert_eq!(
result.usage.as_ref().map(|usage| usage.model.as_str()),
Some("gpt-5.2-1106-preview")
);
Ok(())
}
crates/goose/src/providers/provider_registry.rs
+1
View File
@@ -177,6 +177,7 @@ impl ProviderRegistry {
.iter()
.map(|m| ModelInfo {
name: m.name.clone(),
resolved_model: None,
context_limit: m.context_limit,
input_token_cost: m.input_token_cost,
output_token_cost: m.output_token_cost,
ui/desktop/openapi.json
+33 -1
View File
@@ -5799,6 +5799,25 @@
}
}
},
"InferenceMetadata": {
"type": "object",
"required": [
"provider",
"requestedModel"
],
"properties": {
"provider": {
"type": "string"
},
"requestedModel": {
"type": "string"
},
"resolvedModel": {
"type": "string",
"nullable": true
}
}
},
"InspectJobResponse": {
"type": "object",
"properties": {
@@ -6394,7 +6413,7 @@
},
"MessageMetadata": {
"type": "object",
"description": "Metadata for message visibility",
"description": "Metadata for message visibility and model inference details",
"required": [
"userVisible",
"agentVisible"
@@ -6404,6 +6423,14 @@
"type": "boolean",
"description": "Whether the message should be included in the agent's context window"
},
"inference": {
"allOf": [
{
"$ref": "#/components/schemas/InferenceMetadata"
}
],
"nullable": true
},
"userVisible": {
"type": "boolean",
"description": "Whether the message should be visible to the user in the UI"
@@ -6587,6 +6614,11 @@
"type": "boolean",
"description": "Whether this model supports reasoning/thinking controls"
},
"resolved_model": {
"type": "string",
"description": "The underlying model resolved from provider metadata, when the configured model is an alias or endpoint.",
"nullable": true
},
"supports_cache_control": {
"type": "boolean",
"description": "Whether this model supports cache control",
ui/desktop/src/api/index.ts
+1 -1
View File
File diff suppressed because one or more lines are too long
ui/desktop/src/api/types.gen.ts
+12 -1
View File
@@ -602,6 +602,12 @@ export type ImportSessionRequest = {
json: string;
};
export type InferenceMetadata = {
provider: string;
requestedModel: string;
resolvedModel?: string | null;
};
export type InspectJobResponse = {
processStartTime?: string | null;
runningDurationSeconds?: number | null;
@@ -746,13 +752,14 @@ export type MessageEvent = {
};
/**
* Metadata for message visibility
* Metadata for message visibility and model inference details
*/
export type MessageMetadata = {
/**
* Whether the message should be included in the agent's context window
*/
agentVisible: boolean;
inference?: InferenceMetadata | null;
/**
* Whether the message should be visible to the user in the UI
*/
@@ -822,6 +829,10 @@ export type ModelInfo = {
* Whether this model supports reasoning/thinking controls
*/
reasoning?: boolean;
/**
* The underlying model resolved from provider metadata, when the configured model is an alias or endpoint.
*/
resolved_model?: string | null;
/**
* Whether this model supports cache control
*/
ui/desktop/src/components/BaseChat.tsx
+10
View File
@@ -196,6 +196,15 @@ export default function BaseChat({
const sessionModel = session?.model_config?.model_name ?? null;
const sessionProvider = session?.provider_name ?? null;
const sessionLoaded = session !== undefined;
const latestInference = useMemo(() => {
for (let i = messages.length - 1; i >= 0; i--) {
const message = messages[i];
if (message.role === 'assistant' && message.metadata.userVisible && message.metadata.inference) {
return message.metadata.inference;
}
}
return null;
}, [messages]);
useEffect(() => {
if (!recipe || !isActiveSession) return;
@@ -508,6 +517,7 @@ export default function BaseChat({
sessionModel={sessionModel}
sessionProvider={sessionProvider}
sessionLoaded={sessionLoaded}
latestInference={latestInference}
{...customChatInputProps}
/>
</div>
ui/desktop/src/components/ChatInput.tsx
+3
View File
@@ -204,6 +204,7 @@ interface ChatInputProps {
sessionModel?: string | null;
sessionProvider?: string | null;
sessionLoaded?: boolean;
latestInference?: Message['metadata']['inference'] | null;
}
export default function ChatInput({
@@ -234,6 +235,7 @@ export default function ChatInput({
sessionModel,
sessionProvider,
sessionLoaded,
latestInference,
}: ChatInputProps) {
const [_value, setValue] = useState(initialValue);
const [displayValue, setDisplayValue] = useState(initialValue); // For immediate visual feedback
@@ -1757,6 +1759,7 @@ export default function ChatInput({
setView={setView}
sessionModel={effectiveModel}
sessionProvider={effectiveProvider}
latestInference={latestInference}
onModelChanged={setModelOverride}
sessionLoaded={sessionLoaded}
/>
ui/desktop/src/components/ProgressiveMessageList.tsx
+78 -27
View File
@@ -14,7 +14,7 @@
* - Configurable batch size and delay
*/
import { useCallback, useEffect, useMemo, useRef, useState } from 'react';
import { Fragment, useCallback, useEffect, useMemo, useRef, useState } from 'react';
import { defineMessages, useIntl } from '../i18n';
import { Message, SystemNotificationContent } from '../api';
import GooseMessage from './GooseMessage';
@@ -31,6 +31,7 @@ import { NotificationEvent } from '../types/message';
import LoadingGoose from './LoadingGoose';
import { ChatType } from '../types/chat';
import { identifyConsecutiveToolCalls, isInChain } from '../utils/toolCallChaining';
import { getModelDisplayName } from './settings/models/predefinedModelsUtils';
const i18n = defineMessages({
loadingMessages: {
@@ -41,6 +42,10 @@ const i18n = defineMessages({
id: 'progressiveMessageList.searchHint',
defaultMessage: 'Press Cmd/Ctrl+F to load all messages immediately for search',
},
modelChanged: {
id: 'progressiveMessageList.modelChanged',
defaultMessage: 'Model changed: {previousModel} → {currentModel}',
},
});
interface ProgressiveMessageListProps {
@@ -91,6 +96,37 @@ export default function ProgressiveMessageList({
const hasOnlyToolResponses = (message: Message) =>
message.content.every((c) => c.type === 'toolResponse');
const getResolvedModel = useCallback((message: Message): string | null => {
if (message.role !== 'assistant' || !message.metadata.userVisible) return null;
return message.metadata.inference?.resolvedModel ?? null;
}, []);
const getPreviousResolvedModel = useCallback(
(index: number): string | null => {
for (let i = index - 1; i >= 0; i--) {
const model = getResolvedModel(messages[i]);
if (model) return model;
}
return null;
},
[getResolvedModel, messages]
);
const renderModelChangeDisclosure = useCallback(
(previousModel: string, currentModel: string) => (
<SystemNotificationInline
notification={{
msg: intl.formatMessage(i18n.modelChanged, {
previousModel: getModelDisplayName(previousModel),
currentModel: getModelDisplayName(currentModel),
}),
notificationType: 'inlineMessage',
}}
/>
),
[intl]
);
const getSystemNotification = (message: Message): SystemNotificationContent | undefined => {
return getCreditsExhaustedNotification(message) ?? getInlineSystemNotification(message);
};
@@ -231,34 +267,46 @@ export default function ProgressiveMessageList({
const isUser = isUserMessage(message);
const messageIsInChain = isInChain(index, toolCallChains);
const currentResolvedModel = getResolvedModel(message);
const previousResolvedModel = currentResolvedModel ? getPreviousResolvedModel(index) : null;
const showModelChangeDisclosure = Boolean(
currentResolvedModel &&
previousResolvedModel &&
currentResolvedModel !== previousResolvedModel
);
const messageKey = message.id ?? `msg-${index}-${message.created}`;
return (
<div
key={message.id ?? `msg-${index}-${message.created}`}
className={`relative ${index === 0 ? 'mt-0' : 'mt-4'} ${isUser ? 'user' : 'assistant'} ${messageIsInChain ? 'in-chain' : ''}`}
data-testid="message-container"
>
{isUser ? (
!hasOnlyToolResponses(message) && (
<UserMessage message={message} onMessageUpdate={onMessageUpdate} />
)
) : (
<GooseMessage
sessionId={chat.sessionId}
message={message}
messages={messages}
append={append}
toolCallNotifications={toolCallNotifications}
isStreaming={
isStreamingMessage &&
!isUser &&
index === messagesToRender.length - 1 &&
message.role === 'assistant'
}
submitElicitationResponse={submitElicitationResponse}
/>
)}
</div>
<Fragment key={messageKey}>
{showModelChangeDisclosure && currentResolvedModel && previousResolvedModel &&
renderModelChangeDisclosure(previousResolvedModel, currentResolvedModel)}
<div
className={`relative ${index === 0 ? 'mt-0' : 'mt-4'} ${isUser ? 'user' : 'assistant'} ${messageIsInChain ? 'in-chain' : ''}`}
data-testid="message-container"
>
{isUser ? (
!hasOnlyToolResponses(message) && (
<UserMessage message={message} onMessageUpdate={onMessageUpdate} />
)
) : (
<GooseMessage
sessionId={chat.sessionId}
message={message}
messages={messages}
append={append}
toolCallNotifications={toolCallNotifications}
isStreaming={
isStreamingMessage &&
!isUser &&
index === messagesToRender.length - 1 &&
message.role === 'assistant'
}
submitElicitationResponse={submitElicitationResponse}
/>
)}
</div>
</Fragment>
);
})
.filter(Boolean);
@@ -274,6 +322,9 @@ export default function ProgressiveMessageList({
onMessageUpdate,
toolCallChains,
submitElicitationResponse,
getPreviousResolvedModel,
getResolvedModel,
renderModelChangeDisclosure,
]);
return (
ui/desktop/src/components/settings/models/bottom_bar/ModelsBottomBar.tsx
+29 -1
View File
@@ -1,5 +1,5 @@
import { Sliders, Bot, LoaderCircle, Settings } from 'lucide-react';
import React, { useEffect, useState } from 'react';
import React, { useEffect, useMemo, useState } from 'react';
import { useModelAndProvider } from '../../../ModelAndProviderContext';
import { SwitchModelModal } from '../subcomponents/SwitchModelModal';
import { View } from '../../../../utils/navigationUtils';
@@ -16,6 +16,7 @@ import { getModelDisplayName } from '../predefinedModelsUtils';
import { ModelSettingsPanel } from '../../localInference/ModelSettingsPanel';
import { ScrollArea } from '../../../ui/scroll-area';
import { defineMessages, useIntl } from '../../../../i18n';
import type { Message } from '../../../../api';
const i18n = defineMessages({
selectModel: {
@@ -42,6 +43,10 @@ const i18n = defineMessages({
id: 'modelsBottomBar.localModelSettingsTitle',
defaultMessage: 'Local Model Settings — {modelName}',
},
resolvedModel: {
id: 'modelsBottomBar.resolvedModel',
defaultMessage: 'Resolved model',
},
});
interface ModelsBottomBarProps {
@@ -50,6 +55,7 @@ interface ModelsBottomBarProps {
setView: (view: View) => void;
sessionModel?: string | null;
sessionProvider?: string | null;
latestInference?: Message['metadata']['inference'] | null;
onModelChanged: (override: { model: string; provider: string }) => void;
sessionLoaded?: boolean;
}
@@ -60,6 +66,7 @@ export default function ModelsBottomBar({
setView,
sessionModel,
sessionProvider,
latestInference,
onModelChanged,
sessionLoaded,
}: ModelsBottomBarProps) {
@@ -81,6 +88,14 @@ export default function ModelsBottomBar({
// rather than flashing the config default or leaving the footer blank.
const isModelLoading = Boolean(sessionId && !sessionLoaded);
const displayModel = currentModel || providerDefaultModel || displayModelName;
const resolvedModel = latestInference?.resolvedModel ?? null;
const shouldShowResolvedModel = Boolean(
!isModelLoading &&
resolvedModel &&
latestInference?.provider === currentProvider &&
latestInference?.requestedModel === currentModel &&
resolvedModel !== currentModel
);
const loadingModelLabel = intl.formatMessage(i18n.loadingModel);
const triggerLabel = isModelLoading ? loadingModelLabel : displayModel;
const menuModelLabel = isModelLoading ? loadingModelLabel : displayModelName;
@@ -118,6 +133,11 @@ export default function ModelsBottomBar({
setDisplayModelName(getModelDisplayName(currentModel));
}, [currentModel]);
const resolvedDisplayModelName = useMemo(
() => (resolvedModel ? getModelDisplayName(resolvedModel) : null),
[resolvedModel]
);
const handleModelSelected = (model: string, provider: string) => {
onModelChanged({ model, provider });
};
@@ -147,6 +167,14 @@ export default function ModelsBottomBar({
{menuModelLabel}
{!isModelLoading && displayProvider && ` — ${displayProvider}`}
</p>
{shouldShowResolvedModel && resolvedDisplayModelName && (
<div className="mx-2 pb-2 border-b mb-2">
<h6 className="text-xs text-text-primary">{intl.formatMessage(i18n.resolvedModel)}</h6>
<p className="text-xs text-text-primary truncate" title={resolvedModel ?? undefined}>
{resolvedDisplayModelName}
</p>
</div>
)}
<DropdownMenuItem onClick={() => setIsAddModelModalOpen(true)}>
<span>{intl.formatMessage(i18n.changeModel)}</span>
<Sliders className="ml-auto h-4 w-4 rotate-90" />
ui/desktop/src/hooks/useChatStream.ts
+55 -1
View File
@@ -183,6 +183,13 @@ function pushMessage(currentMessages: Message[], incomingMsg: Message): Message[
const lastContent = lastMsg.content[lastMsg.content.length - 1];
const newContent = incomingMsg.content[incomingMsg.content.length - 1];
if (incomingMsg.metadata?.inference) {
lastMsg.metadata = {
...lastMsg.metadata,
inference: incomingMsg.metadata.inference,
};
}
if (
lastContent?.type === 'text' &&
newContent?.type === 'text' &&
@@ -236,6 +243,7 @@ function createEventProcessor(
let latestChatState: ChatState = ChatState.Streaming;
let lastBatchUpdate = Date.now();
let hasPendingUpdate = false;
let pendingInference: Message['metadata']['inference'] | undefined;
const flushBatchedUpdates = () => {
if (reduceMotion && hasPendingUpdate) {
@@ -271,12 +279,54 @@ function createEventProcessor(
}
};
const flushPendingInference = () => {
if (!pendingInference) {
return;
}
for (let i = currentMessages.length - 1; i >= 0; i--) {
const message = currentMessages[i];
if (message.role === 'assistant' && message.metadata.userVisible) {
currentMessages = [
...currentMessages.slice(0, i),
{
...message,
metadata: {
...message.metadata,
inference: message.metadata.inference ?? pendingInference,
},
},
...currentMessages.slice(i + 1),
];
break;
}
}
pendingInference = undefined;
};
// Returns true if the event is terminal (Finish or Error)
const processEvent = (event: SessionEvent): boolean => {
switch (event.type) {
case 'Message': {
const msg = (event as Record<string, unknown>).message as Message;
let msg = (event as Record<string, unknown>).message as Message;
const tokenState = (event as Record<string, unknown>).token_state as TokenState;
if (msg.content.length === 0 && msg.metadata?.inference) {
pendingInference = msg.metadata.inference;
return false;
}
if (pendingInference && msg.role === 'assistant' && msg.metadata.userVisible) {
msg = {
...msg,
metadata: {
...msg.metadata,
inference: msg.metadata.inference ?? pendingInference,
},
};
pendingInference = undefined;
}
currentMessages = pushMessage(currentMessages, msg);
const hasToolConfirmation = msg.content.some(
@@ -301,7 +351,9 @@ function createEventProcessor(
return false;
}
case 'Error': {
flushPendingInference();
flushBatchedUpdates();
dispatch({ type: 'SET_MESSAGES', payload: currentMessages });
const errorMsg = String((event as Record<string, unknown>).error ?? '');
if (errorMsg.includes('too far behind') && onReloadNeeded) {
// Server indicated we missed events — end streaming without setting
@@ -315,7 +367,9 @@ function createEventProcessor(
return true;
}
case 'Finish': {
flushPendingInference();
flushBatchedUpdates();
dispatch({ type: 'SET_MESSAGES', payload: currentMessages });
onFinish();
return true;
}
ui/desktop/src/i18n/messages/en.json
+6
View File
@@ -2390,6 +2390,9 @@
"modelsBottomBar.localModelSettingsTitle": {
"defaultMessage": "Local Model Settings — {modelName}"
},
"modelsBottomBar.resolvedModel": {
"defaultMessage": "Resolved model"
},
"modelsBottomBar.selectModel": {
"defaultMessage": "Select Model"
},
@@ -2687,6 +2690,9 @@
"progressiveMessageList.loadingMessages": {
"defaultMessage": "Loading messages... ({renderedCount}/{totalCount})"
},
"progressiveMessageList.modelChanged": {
"defaultMessage": "Model changed: {previousModel} → {currentModel}"
},
"progressiveMessageList.searchHint": {
"defaultMessage": "Press Cmd/Ctrl+F to load all messages immediately for search"
},
ui/desktop/src/i18n/messages/zh-CN.json
+6
View File
@@ -2333,6 +2333,9 @@
"modelsBottomBar.localModelSettingsTitle": {
"defaultMessage": "本地模型设置 — {modelName}"
},
"modelsBottomBar.resolvedModel": {
"defaultMessage": "解析后的模型"
},
"modelsBottomBar.selectModel": {
"defaultMessage": "选择模型"
},
@@ -2630,6 +2633,9 @@
"progressiveMessageList.loadingMessages": {
"defaultMessage": "正在加载消息…（{renderedCount}/{totalCount}）"
},
"progressiveMessageList.modelChanged": {
"defaultMessage": "模型已更改：{previousModel} → {currentModel}"
},
"progressiveMessageList.searchHint": {
"defaultMessage": "按 Cmd/Ctrl+F 立即加载全部消息以便搜索"
},