c3d31c06e2
- Auth/roles (no self-reg), admin user provision, JWT - Analyze: sales kit + initial pain-fit from form/upload - Persona generator: 15 personas (5/tier) w/ pain variety, negotiation, init mode, channel, latent/revealable, wrong_text special - Chat simulator: per-mode initiation, one-shot, hidden signals, judge-LLM debrief+coaching - Trainee loop: win/lose board, weak-areas, user-generated personas - Admin analytics; EN+TH Vue SPA served by Flask - Deploy: Dockerfile, docker-compose, README, eng-log + HANDOFF - Tests (mock LLM): m0/m1/routes/e2e all pass
132 lines
4.0 KiB
Python
132 lines
4.0 KiB
Python
"""OpenAI-compatible LLM client (OpenAI / DeepSeek / custom base URL).
|
|
|
|
Mirrors the MiroFish provider-agnostic pattern. Credentials live in .env only.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import re
|
|
from typing import Any
|
|
|
|
from openai import OpenAI
|
|
|
|
from .config import Config
|
|
|
|
|
|
class LLMError(Exception):
|
|
pass
|
|
|
|
|
|
def _strip_thinking_trace(text: str) -> str:
|
|
"""Remove ReACT-style chain-of-thought / fences, keep the final JSON text."""
|
|
for fence in ("```json", "```"):
|
|
idx = text.rfind(fence)
|
|
if idx != -1:
|
|
after = text[idx:].lstrip()
|
|
lang_len = after.find("\n")
|
|
body = after[lang_len:] if lang_len != -1 else after
|
|
end = body.rfind("```")
|
|
if end != -1:
|
|
body = body[:end]
|
|
body = body.strip()
|
|
if body:
|
|
return body
|
|
for marker in ("\n\n[", "\n\n{"):
|
|
idx = text.rfind(marker)
|
|
if idx != -1:
|
|
candidate = text[idx:].strip()
|
|
if candidate and candidate[0] in "{[":
|
|
return candidate
|
|
return text
|
|
|
|
|
|
class LLMClient:
|
|
def __init__(
|
|
self,
|
|
*,
|
|
base_url: str | None = None,
|
|
api_key: str | None = None,
|
|
model: str | None = None,
|
|
) -> None:
|
|
self.base_url = base_url or Config.LLM_BASE_URL
|
|
self.api_key = api_key or Config.LLM_API_KEY
|
|
self.model = model or Config.LLM_MODEL
|
|
if not self.api_key:
|
|
raise LLMError("LLM_API_KEY is not configured in .env")
|
|
if not self.base_url:
|
|
raise LLMError("LLM_BASE_URL is not configured (unknown provider)")
|
|
self.client = OpenAI(base_url=self.base_url, api_key=self.api_key)
|
|
|
|
def complete(
|
|
self,
|
|
system_prompt: str,
|
|
user_prompt: str,
|
|
*,
|
|
temperature: float = 0.5,
|
|
max_tokens: int = 3000,
|
|
) -> str:
|
|
try:
|
|
resp = self.client.chat.completions.create(
|
|
model=self.model,
|
|
temperature=temperature,
|
|
max_tokens=max_tokens,
|
|
messages=[
|
|
{"role": "system", "content": system_prompt},
|
|
{"role": "user", "content": user_prompt},
|
|
],
|
|
)
|
|
except Exception as exc: # network/auth/provider
|
|
raise LLMError(f"LLM call failed: {exc}") from exc
|
|
text = (resp.choices[0].message.content or "").strip()
|
|
if not text:
|
|
raise LLMError("LLM returned empty response")
|
|
return text
|
|
|
|
def complete_json(
|
|
self,
|
|
system_prompt: str,
|
|
user_prompt: str,
|
|
*,
|
|
temperature: float = 0.2,
|
|
max_tokens: int = 6000,
|
|
) -> dict[str, Any]:
|
|
text = self.complete(
|
|
system_prompt,
|
|
user_prompt,
|
|
temperature=temperature,
|
|
max_tokens=max_tokens,
|
|
)
|
|
text = _strip_thinking_trace(text)
|
|
try:
|
|
return json.loads(text)
|
|
except json.JSONDecodeError as exc:
|
|
# Last-ditch: strip leading text before the first { or [
|
|
match = re.search(r"[{\[].*[}\]]", text, re.DOTALL)
|
|
if match:
|
|
try:
|
|
return json.loads(match.group(0))
|
|
except json.JSONDecodeError:
|
|
pass
|
|
raise LLMError(f"LLM returned invalid JSON: {exc}") from exc
|
|
|
|
def complete_conversation(
|
|
self,
|
|
messages: list[dict[str, str]],
|
|
*,
|
|
temperature: float = 0.6,
|
|
max_tokens: int = 1200,
|
|
) -> str:
|
|
try:
|
|
resp = self.client.chat.completions.create(
|
|
model=self.model,
|
|
temperature=temperature,
|
|
max_tokens=max_tokens,
|
|
messages=messages,
|
|
)
|
|
except Exception as exc:
|
|
raise LLMError(f"LLM call failed: {exc}") from exc
|
|
text = (resp.choices[0].message.content or "").strip()
|
|
if not text:
|
|
raise LLMError("LLM returned empty response")
|
|
return text
|