Skip to content

Fix: decouple chat from local Ollama, route inference through Rust #369

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

Merged
senamakel merged 8 commits into from
Apr 6, 2026
Merged

Fix: decouple chat from local Ollama, route inference through Rust #369

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
912a790
refactor: streamline Conversations component by removing local respon…
senamakel Apr 6, 2026
2906587
refactor: update chat service to utilize core RPC for message handling
senamakel Apr 6, 2026
acb6ee9
feat: enhance chat response handling with segmentation and emoji reac…
senamakel Apr 6, 2026
3a66e9b
refactor: simplify Conversations component by removing unused imports…
senamakel Apr 6, 2026
422a424
style: apply formatter fixes (prettier + cargo fmt)
senamakel Apr 6, 2026
0f9d8bb
updated convos
senamakel Apr 6, 2026
a38062b
Merge remote-tracking branch 'upstream/main' into fix/ollama-2
senamakel Apr 6, 2026
ec718e8
fix test
senamakel Apr 6, 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
230 changes: 49 additions & 181 deletions app/src/pages/Conversations.tsx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,13 +4,14 @@ import Markdown from 'react-markdown';
import { useNavigate } from 'react-router-dom';

import { type ChatSendError, chatSendError } from '../chat/chatSendError';
import { useLocalModelStatus } from '../hooks/useLocalModelStatus';
import { creditsApi, type TeamUsage } from '../services/api/creditsApi';
import {
chatCancel,
type ChatSegmentEvent,
chatSend,
type ChatToolCallEvent,
type ChatToolResultEvent,
segmentText,
subscribeChatEvents,
useRustChat,
} from '../services/chatService';
Expand All @@ -28,14 +29,10 @@ import {
setSelectedThread,
} from '../store/threadSlice';
import type { ThreadMessage } from '../types/thread';
import { getSegmentDelay, segmentMessage } from '../utils/messageSegmentation';
import {
isTauri,
type LocalAiChatMessage,
openhumanAutocompleteAccept,
openhumanAutocompleteCurrent,
openhumanLocalAiChat,
openhumanLocalAiShouldReact,
openhumanVoiceStatus,
openhumanVoiceTranscribeBytes,
openhumanVoiceTts,
Expand Down Expand Up @@ -141,14 +138,7 @@ const Conversations = () => {
Record<string, ToolTimelineEntry[]>
>({});
const rustChat = useRustChat();
const isLocalModelActive = useLocalModelStatus();
const isLocalModelActiveRef = useRef(isLocalModelActive);
const [isDelivering, setIsDelivering] = useState(false);
const deliveryActiveRef = useRef(false);
const [reactionPickerMsgId, setReactionPickerMsgId] = useState<string | null>(null);
const defaultChannelType = useAppSelector(
state => state.channelConnections?.defaultMessagingChannel ?? 'web'
);
const pendingReactionRef = useRef<
Map<string, { msgId: string; content: string; threadId: string }>
>(new Map());
Expand All @@ -158,10 +148,6 @@ const Conversations = () => {
selectedThreadIdRef.current = selectedThreadId;
}, [selectedThreadId]);

useEffect(() => {
isLocalModelActiveRef.current = isLocalModelActive;
}, [isLocalModelActive]);

const [teamUsage, setTeamUsage] = useState<TeamUsage | null>(null);
const [isLoadingBudget, setIsLoadingBudget] = useState(false);

Expand Down Expand Up @@ -305,7 +291,6 @@ const Conversations = () => {

useEffect(() => {
return () => {
deliveryActiveRef.current = false;
mediaRecorderRef.current?.stop();
mediaStreamRef.current?.getTracks().forEach(track => track.stop());
replyAudioRef.current?.pause();
Expand Down Expand Up @@ -399,16 +384,24 @@ const Conversations = () => {
return { ...prev, [event.thread_id]: nextEntries };
});
},
onDone: event => {
const currentState = store.getState() as {
thread: { messagesByThreadId: Record<string, ThreadMessage[]> };
};
const threadMessages = currentState.thread.messagesByThreadId[event.thread_id] || [];
const lastMsg = threadMessages[threadMessages.length - 1];
if (lastMsg?.sender === 'agent' && lastMsg?.content === event.full_response) {
return;
onSegment: (event: ChatSegmentEvent) => {
// Rust delivers segments with delays already applied — just dispatch.
if (event.reaction_emoji) {
const pending = pendingReactionRef.current.get(event.thread_id);
if (pending) {
dispatch(
addReaction({
threadId: pending.threadId,
messageId: pending.msgId,
emoji: event.reaction_emoji,
})
);
pendingReactionRef.current.delete(event.thread_id);
}
}

dispatch(addInferenceResponse({ content: segmentText(event), threadId: event.thread_id }));
},
onDone: event => {
// Update tool timeline
setToolTimelineByThread(prev => {
const existing = prev[event.thread_id] ?? [];
Expand All @@ -425,58 +418,31 @@ const Conversations = () => {
sendingTimeoutRef.current = null;
}

// Fire-and-forget: auto-react to the user's message
const pending = pendingReactionRef.current.get(event.thread_id);
if (pending) {
maybeAutoReact(pending.msgId, pending.content, pending.threadId);
pendingReactionRef.current.delete(event.thread_id);
}

// Multi-bubble delivery gate: only when local model is active
if (!isLocalModelActiveRef.current) {
dispatch(
addInferenceResponse({ content: event.full_response, threadId: event.thread_id })
);
setIsSending(false);
dispatch(setActiveThread(null));
return;
// Apply reaction emoji from Rust (when not segmented — no onSegment fired).
if (event.reaction_emoji) {
const pending = pendingReactionRef.current.get(event.thread_id);
if (pending) {
dispatch(
addReaction({
threadId: pending.threadId,
messageId: pending.msgId,
emoji: event.reaction_emoji,
})
);
}
}
pendingReactionRef.current.delete(event.thread_id);

const segments = segmentMessage(event.full_response);

if (segments.length <= 1) {
// Only add the response bubble if Rust didn't already deliver it
// via chat_segment events (segment_total > 0 means segments were sent).
if (!event.segment_total) {
dispatch(
addInferenceResponse({ content: event.full_response, threadId: event.thread_id })
);
setIsSending(false);
dispatch(setActiveThread(null));
return;
}

// Async delivery: show each segment as a separate bubble with a typing pause
setIsDelivering(true);
deliveryActiveRef.current = true;

void (async () => {
for (let i = 0; i < segments.length; i++) {
if (!deliveryActiveRef.current) break;

if (i > 0) {
await new Promise<void>(resolve =>
setTimeout(resolve, getSegmentDelay(segments[i - 1]))
);
}

if (!deliveryActiveRef.current) break;

dispatch(addInferenceResponse({ content: segments[i], threadId: event.thread_id }));
}

deliveryActiveRef.current = false;
setIsDelivering(false);
setIsSending(false);
// activeThreadId was already cleared by the first addInferenceResponse dispatch
})();
setIsSending(false);
dispatch(setActiveThread(null));
},
onError: event => {
if (event.thread_id !== selectedThreadIdRef.current) return;
Expand Down Expand Up @@ -529,65 +495,18 @@ const Conversations = () => {
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [rustChat, socketStatus]);

/**
* Segment a complete response string and dispatch each segment as a
* separate message bubble with a typing pause between them.
* Local-model-only path — no cloud API calls.
*/
const deliverLocalResponse = async (fullResponse: string, threadId: string) => {
const segments = segmentMessage(fullResponse);

if (segments.length <= 1) {
dispatch(addInferenceResponse({ content: fullResponse, threadId }));
return;
}

setIsDelivering(true);
deliveryActiveRef.current = true;

for (let i = 0; i < segments.length; i++) {
if (!deliveryActiveRef.current) break;

if (i > 0) {
await new Promise<void>(resolve => setTimeout(resolve, getSegmentDelay(segments[i - 1])));
}

if (!deliveryActiveRef.current) break;

dispatch(addInferenceResponse({ content: segments[i], threadId }));
}

deliveryActiveRef.current = false;
setIsDelivering(false);
};

/**
* Fire-and-forget: ask the local model if we should auto-react to the
* user's message with an emoji. Adds a personal touch based on channel type.
*/
const maybeAutoReact = (userMessageId: string, messageContent: string, threadId: string) => {
if (!isTauri() || !isLocalModelActiveRef.current) return;

void openhumanLocalAiShouldReact(messageContent, defaultChannelType)
.then(response => {
const decision = response.result;
if (decision?.should_react && decision.emoji) {
console.debug('[conversations:auto-react] reacting with', decision.emoji);
dispatch(addReaction({ threadId, messageId: userMessageId, emoji: decision.emoji }));
}
})
.catch(err => {
console.debug('[conversations:auto-react] failed:', err);
});
};

const handleSendMessage = async (text?: string) => {
const normalized = text ?? inputValue;
const trimmed = normalized.trim();

if (!trimmed || !selectedThreadId || isSending) return;
if (!isLocalModelActiveRef.current && socketStatus !== 'connected') {
setSendError(chatSendError('socket_disconnected', 'Realtime socket is not connected.'));
if (socketStatus !== 'connected') {
setSendError(
chatSendError(
'socket_disconnected',
'Realtime socket is not connected — responses cannot be delivered without a client ID.'
)
);
return;
}

Expand Down Expand Up @@ -633,61 +552,10 @@ const Conversations = () => {
setToolTimelineByThread(prev => ({ ...prev, [sendingThreadId]: [] }));
dispatch(setActiveThread(sendingThreadId));

// ── Local Ollama path ────────────────────────────────────────────────────
// When a local model is ready, bypass the cloud socket entirely.
// Zero cloud tokens consumed on this path.
if (isLocalModelActiveRef.current) {
try {
// Build message history: convert stored messages + the new user turn
const storedMessages =
(
store.getState() as {
thread: {
messagesByThreadId: Record<string, import('../types/thread').ThreadMessage[]>;
};
}
).thread.messagesByThreadId[sendingThreadId] ?? [];

const history: LocalAiChatMessage[] = storedMessages
.filter(m => m.sender === 'user' || m.sender === 'agent')
.map(m => ({
role: m.sender === 'user' ? ('user' as const) : ('assistant' as const),
content: m.content,
}));

console.debug('[conversations:local] sending to local model', {
historyLength: history.length,
threadId: sendingThreadId,
});

const response = await openhumanLocalAiChat(history);
const reply = response.result?.trim() ?? '';

if (!reply) {
throw new Error('Local model returned an empty response.');
}

await deliverLocalResponse(reply, sendingThreadId);
pendingReactionRef.current.delete(sendingThreadId);
maybeAutoReact(userMessage.id, trimmed, sendingThreadId);
} catch (err) {
pendingReactionRef.current.delete(sendingThreadId);
const msg = err instanceof Error ? err.message : String(err);
setSendError(chatSendError('local_model_failed', msg));
dispatch(
addInferenceResponse({
content: 'Local model error — please try again.',
threadId: sendingThreadId,
})
);
} finally {
setIsSending(false);
dispatch(setActiveThread(null));
}
return;
}

// ── Cloud socket path (unchanged) ────────────────────────────────────────
// ── Cloud socket path ─────────────────────────────────────────────────────
// Always route primary chat through the cloud backend via socket.
// Local model (Ollama) is used only for supplementary features
// (auto-react, autocomplete, etc.) — never as a primary chat path.
try {
await chatSend({ threadId: sendingThreadId, message: trimmed, model: AGENTIC_MODEL_ID });

Expand Down Expand Up @@ -1082,7 +950,7 @@ const Conversations = () => {
</div>
</div>
))}
{((activeThreadId === selectedThreadId && isSending) || isDelivering) && (
{activeThreadId === selectedThreadId && isSending && (
<div className="flex justify-start">
<div className="bg-stone-200/80 rounded-2xl rounded-bl-md px-4 py-3">
<div className="flex items-center gap-1">
Expand Down
Loading
Loading