Files
Heimgeist/backend/ollama_client.py

59 lines
2.2 KiB
Python
Raw Normal View History

2025-08-22 23:42:34 +02:00
import httpx
feat: Add streaming chat + scroll persistence; improve markdown & links Backend - /chat: support streaming via StreamingResponse; save full reply after stream ends. Non-stream path unchanged. - ChatRequest: add stream flag (default false). - GenerateTitleRequest: add model and use it instead of hardcoded llama3. - ollama_client.chat_stream(): new async generator parsing Ollama streaming JSON (both formats). - Remove response_model from /chat to allow streaming; non-stream still returns { reply }. Electron - Open external links in system browser (setWindowOpenHandler, shell.openExternal). - New IPC: update-settings, open-external-link. - Set minimum window size; preload exposes updateSettings and openExternalLink. Frontend (React) - Streaming UI with live chunking; sticky-bottom only when user at bottom. - Per-session scroll persistence and robust restore. - New message tip to jump to latest reply when scrolled up. - Disable Send while sending; spinner. - General Settings: stream output toggle; propagate model/stream changes. - Apply color scheme at boot; extract colorSchemes helper. - Sidebar UX tweaks and unread badges. Markdown/rendering - Code blocks: language title bar and wrapper. - Tables: GitHub-style parsing, per-cell borders, rounded wrapper, spacing, alignment. - Headings: remove blank line after h1-h4. - <hr>: handle after tables; strip following whitespace. - Links: target=_blank with icon and URL tooltip. Styles - Add styles for code/table wrappers, new-message tip, toggle, spinner; hover/active vars; narrower sidebar. API notes / breaking changes - /chat accepts stream=true and returns text/plain streamed chunks. - generate-title now requires a model. - Non-stream /chat response shape unchanged.
2025-08-23 16:45:46 +02:00
import json
from typing import Dict, Any, List, AsyncGenerator
2025-08-22 23:42:34 +02:00
from .app_settings import get_ollama_api_url
2025-08-22 23:42:34 +02:00
async def list_models() -> Dict[str, Any]:
ollama_url = get_ollama_api_url()
2025-08-22 23:42:34 +02:00
async with httpx.AsyncClient(timeout=30.0) as client:
r = await client.get(f"{ollama_url}/api/tags")
2025-08-22 23:42:34 +02:00
r.raise_for_status()
data = r.json()
# Normalize to a simple list of names
models = [m.get('name') for m in data.get('models', [])]
return {"models": models}
async def chat(model: str, messages: List[Dict[str, str]]) -> str:
ollama_url = get_ollama_api_url()
2025-08-22 23:42:34 +02:00
payload = {
"model": model,
"messages": messages,
"stream": False
}
async with httpx.AsyncClient(timeout=600.0) as client:
r = await client.post(f"{ollama_url}/api/chat", json=payload)
2025-08-22 23:42:34 +02:00
r.raise_for_status()
data = r.json()
# Ollama returns full conversation; pick last message content
try:
return data["message"]["content"]
except Exception:
# Newer Ollama formats may return messages list
msgs = data.get("messages") or []
if msgs:
return msgs[-1].get("content", "")
return data.get("content", "")
feat: Add streaming chat + scroll persistence; improve markdown & links Backend - /chat: support streaming via StreamingResponse; save full reply after stream ends. Non-stream path unchanged. - ChatRequest: add stream flag (default false). - GenerateTitleRequest: add model and use it instead of hardcoded llama3. - ollama_client.chat_stream(): new async generator parsing Ollama streaming JSON (both formats). - Remove response_model from /chat to allow streaming; non-stream still returns { reply }. Electron - Open external links in system browser (setWindowOpenHandler, shell.openExternal). - New IPC: update-settings, open-external-link. - Set minimum window size; preload exposes updateSettings and openExternalLink. Frontend (React) - Streaming UI with live chunking; sticky-bottom only when user at bottom. - Per-session scroll persistence and robust restore. - New message tip to jump to latest reply when scrolled up. - Disable Send while sending; spinner. - General Settings: stream output toggle; propagate model/stream changes. - Apply color scheme at boot; extract colorSchemes helper. - Sidebar UX tweaks and unread badges. Markdown/rendering - Code blocks: language title bar and wrapper. - Tables: GitHub-style parsing, per-cell borders, rounded wrapper, spacing, alignment. - Headings: remove blank line after h1-h4. - <hr>: handle after tables; strip following whitespace. - Links: target=_blank with icon and URL tooltip. Styles - Add styles for code/table wrappers, new-message tip, toggle, spinner; hover/active vars; narrower sidebar. API notes / breaking changes - /chat accepts stream=true and returns text/plain streamed chunks. - generate-title now requires a model. - Non-stream /chat response shape unchanged.
2025-08-23 16:45:46 +02:00
async def chat_stream(model: str, messages: List[Dict[str, str]]) -> AsyncGenerator[str, None]:
ollama_url = get_ollama_api_url()
feat: Add streaming chat + scroll persistence; improve markdown & links Backend - /chat: support streaming via StreamingResponse; save full reply after stream ends. Non-stream path unchanged. - ChatRequest: add stream flag (default false). - GenerateTitleRequest: add model and use it instead of hardcoded llama3. - ollama_client.chat_stream(): new async generator parsing Ollama streaming JSON (both formats). - Remove response_model from /chat to allow streaming; non-stream still returns { reply }. Electron - Open external links in system browser (setWindowOpenHandler, shell.openExternal). - New IPC: update-settings, open-external-link. - Set minimum window size; preload exposes updateSettings and openExternalLink. Frontend (React) - Streaming UI with live chunking; sticky-bottom only when user at bottom. - Per-session scroll persistence and robust restore. - New message tip to jump to latest reply when scrolled up. - Disable Send while sending; spinner. - General Settings: stream output toggle; propagate model/stream changes. - Apply color scheme at boot; extract colorSchemes helper. - Sidebar UX tweaks and unread badges. Markdown/rendering - Code blocks: language title bar and wrapper. - Tables: GitHub-style parsing, per-cell borders, rounded wrapper, spacing, alignment. - Headings: remove blank line after h1-h4. - <hr>: handle after tables; strip following whitespace. - Links: target=_blank with icon and URL tooltip. Styles - Add styles for code/table wrappers, new-message tip, toggle, spinner; hover/active vars; narrower sidebar. API notes / breaking changes - /chat accepts stream=true and returns text/plain streamed chunks. - generate-title now requires a model. - Non-stream /chat response shape unchanged.
2025-08-23 16:45:46 +02:00
payload = {
"model": model,
"messages": messages,
"stream": True
}
async with httpx.AsyncClient(timeout=600.0) as client:
async with client.stream("POST", f"{ollama_url}/api/chat", json=payload) as r:
feat: Add streaming chat + scroll persistence; improve markdown & links Backend - /chat: support streaming via StreamingResponse; save full reply after stream ends. Non-stream path unchanged. - ChatRequest: add stream flag (default false). - GenerateTitleRequest: add model and use it instead of hardcoded llama3. - ollama_client.chat_stream(): new async generator parsing Ollama streaming JSON (both formats). - Remove response_model from /chat to allow streaming; non-stream still returns { reply }. Electron - Open external links in system browser (setWindowOpenHandler, shell.openExternal). - New IPC: update-settings, open-external-link. - Set minimum window size; preload exposes updateSettings and openExternalLink. Frontend (React) - Streaming UI with live chunking; sticky-bottom only when user at bottom. - Per-session scroll persistence and robust restore. - New message tip to jump to latest reply when scrolled up. - Disable Send while sending; spinner. - General Settings: stream output toggle; propagate model/stream changes. - Apply color scheme at boot; extract colorSchemes helper. - Sidebar UX tweaks and unread badges. Markdown/rendering - Code blocks: language title bar and wrapper. - Tables: GitHub-style parsing, per-cell borders, rounded wrapper, spacing, alignment. - Headings: remove blank line after h1-h4. - <hr>: handle after tables; strip following whitespace. - Links: target=_blank with icon and URL tooltip. Styles - Add styles for code/table wrappers, new-message tip, toggle, spinner; hover/active vars; narrower sidebar. API notes / breaking changes - /chat accepts stream=true and returns text/plain streamed chunks. - generate-title now requires a model. - Non-stream /chat response shape unchanged.
2025-08-23 16:45:46 +02:00
r.raise_for_status()
async for line in r.aiter_lines():
if line:
try:
chunk = json.loads(line)
if "content" in chunk: # Newer Ollama format
yield chunk["content"]
elif "message" in chunk and "content" in chunk["message"]: # Older format
yield chunk["message"]["content"]
except json.JSONDecodeError:
pass # Ignore invalid JSON lines