Merge pull request 'feat(vue): open Vue SPA to all tiers; fix cloud navigation and feedback button' (#64) from feature/vue-streamlit-parity into main

- Lower vue_ui_beta gate to "free" so all licensed users can access the
  new UI without a paid subscription
- Remove "Paid tier" wording from the Try New UI banner
- Fix Vue SPA navigation in cloud/demo deployments: add VITE_BASE_PATH
  build arg so Vite sets the correct subpath base, and pass
  import.meta.env.BASE_URL to createWebHistory() so router links
  emit /peregrine/... paths that Caddy can match
- Fix feedback button missing on cloud instance by passing
  FORGEJO_API_TOKEN through compose.cloud.yml
- Remove vLLM container from compose.yml (vLLM dropped from stack;
  cf-research service in cfcore covers the use case)
- Fix cloud config path in Apply page (use get_config_dir() so per-user
  cloud data roots resolve correctly for user.yaml and resume YAML)
- Refactor generate_cover_letter._build_system_context and
  _build_mission_notes to accept explicit profile arg (enables
  per-user cover letter generation in cloud multi-tenant mode)
- Add API proxy block to nginx.conf (Vue web container can now call
  /api/ directly without Vite dev proxy)
- Update .env.example: remove vLLM vars, add research model + tuning
  vars for external vLLM deployments
- Update llm.yaml: switch vllm base_url to host.docker.internal
  (vLLM now runs outside Docker stack)

Closes #63 (feedback button)
Related: #8 (Vue SPA), #50–#62 (parity milestone)

.env.example

@@ -12,8 +12,11 @@ VISION_REVISION=2025-01-09
 
 DOCS_DIR=~/Documents/JobSearch
 OLLAMA_MODELS_DIR=~/models/ollama
-VLLM_MODELS_DIR=~/models/vllm
+VLLM_MODELS_DIR=~/models/vllm        # override with full path to your model dir
-VLLM_MODEL=Ouro-1.4B
+VLLM_MODEL=Ouro-1.4B                # cover letters — fast 1.4B model
+VLLM_RESEARCH_MODEL=Ouro-2.6B-Thinking  # research — reasoning 2.6B model; restart vllm to switch
+VLLM_MAX_MODEL_LEN=4096             # increase to 8192 for Thinking models with long CoT
+VLLM_GPU_MEM_UTIL=0.75              # lower to 0.6 if sharing GPU with other services
 OLLAMA_DEFAULT_MODEL=llama3.2:3b
 
 # API keys (required for remote profile)

app/components/ui_switcher.py

@@ -200,7 +200,7 @@ def render_banner(yaml_path: Path, tier: str) -> None:
 
     col1, col2, col3 = st.columns([8, 1, 1])
     with col1:
-        st.info("✨ **New Peregrine UI available** — try the modern Vue interface (Beta, Paid tier)")
+        st.info("✨ **New Peregrine UI available** — try the modern Vue interface (Beta)")
     with col2:
         if st.button("Try it", key="_ui_banner_try"):
             switch_ui(yaml_path, to="vue", tier=tier)

app/pages/4_Apply.py

@@ -15,28 +15,28 @@ import streamlit.components.v1 as components
 import yaml
 
 from scripts.user_profile import UserProfile
-
-_USER_YAML = Path(__file__).parent.parent.parent / "config" / "user.yaml"
-_profile = UserProfile(_USER_YAML) if UserProfile.exists(_USER_YAML) else None
-_name = _profile.name if _profile else "Job Seeker"
-
 from scripts.db import (
     DEFAULT_DB, init_db, get_jobs_by_status,
     update_cover_letter, mark_applied, update_job_status,
     get_task_for_job,
 )
 from scripts.task_runner import submit_task
-from app.cloud_session import resolve_session, get_db_path
+from app.cloud_session import resolve_session, get_db_path, get_config_dir
 from app.telemetry import log_usage_event
 
-DOCS_DIR = _profile.docs_dir if _profile else Path.home() / "Documents" / "JobSearch"
-RESUME_YAML = Path(__file__).parent.parent.parent / "config" / "plain_text_resume.yaml"
-
 st.title("🚀 Apply Workspace")
 
 resolve_session("peregrine")
 init_db(get_db_path())
 
+_CONFIG_DIR = get_config_dir()
+_USER_YAML = _CONFIG_DIR / "user.yaml"
+_profile = UserProfile(_USER_YAML) if UserProfile.exists(_USER_YAML) else None
+_name = _profile.name if _profile else "Job Seeker"
+
+DOCS_DIR = _profile.docs_dir if _profile else Path.home() / "Documents" / "JobSearch"
+RESUME_YAML = _CONFIG_DIR / "plain_text_resume.yaml"
+
 # ── PDF generation ─────────────────────────────────────────────────────────────
 def _make_cover_letter_pdf(job: dict, cover_letter: str, output_dir: Path) -> Path:
     from reportlab.lib.pagesizes import letter

app/wizard/tiers.py

@@ -64,8 +64,8 @@ FEATURES: dict[str, str] = {
     "apple_calendar_sync":          "paid",
     "slack_notifications":          "paid",
 
-    # Beta UI access — stays gated (access management, not compute)
+    # Beta UI access — open to all tiers (access management, not compute)
-    "vue_ui_beta":                  "paid",
+    "vue_ui_beta":                  "free",
 }
 
 # Features that unlock when the user supplies any LLM backend (local or BYOK).

compose.cloud.yml

@@ -37,6 +37,7 @@ services:
       - PEREGRINE_CADDY_PROXY=1
       - CF_ORCH_URL=http://host.docker.internal:7700
       - DEMO_MODE=false
+      - FORGEJO_API_TOKEN=${FORGEJO_API_TOKEN:-}
     depends_on:
       searxng:
         condition: service_healthy
@@ -48,6 +49,8 @@ services:
     build:
       context: .
       dockerfile: docker/web/Dockerfile
+      args:
+        VITE_BASE_PATH: /peregrine/
     ports:
       - "8508:80"
     restart: unless-stopped

compose.demo.yml

@@ -42,6 +42,8 @@ services:
     build:
       context: .
       dockerfile: docker/web/Dockerfile
+      args:
+        VITE_BASE_PATH: /peregrine/
     ports:
       - "8507:80"
     restart: unless-stopped

compose.yml

@@ -1,5 +1,5 @@
 # compose.yml — Peregrine by Circuit Forge LLC
-# Profiles: remote | cpu | single-gpu | dual-gpu-ollama | dual-gpu-vllm | dual-gpu-mixed
+# Profiles: remote | cpu | single-gpu | dual-gpu-ollama
 services:
 
   app:
@@ -129,23 +129,6 @@ services:
     profiles: [single-gpu, dual-gpu-ollama, dual-gpu-vllm, dual-gpu-mixed]
     restart: unless-stopped
 
-  vllm:
-    image: vllm/vllm-openai:latest
-    ports:
-      - "${VLLM_PORT:-8000}:8000"
-    volumes:
-      - ${VLLM_MODELS_DIR:-~/models/vllm}:/models
-    command: >
-      --model /models/${VLLM_MODEL:-Ouro-1.4B}
-      --trust-remote-code
-      --max-model-len 4096
-      --gpu-memory-utilization 0.75
-      --enforce-eager
-      --max-num-seqs 8
-      --cpu-offload-gb ${CPU_OFFLOAD_GB:-0}
-    profiles: [dual-gpu-vllm, dual-gpu-mixed]
-    restart: unless-stopped
-
   finetune:
     build:
       context: .

config/llm.yaml

@@ -28,9 +28,9 @@ backends:
     type: openai_compat
   ollama_research:
     api_key: ollama
-    base_url: http://host.docker.internal:11434/v1
+    base_url: http://ollama_research:11434/v1
     enabled: true
-    model: llama3.2:3b
+    model: llama3.1:8b
     supports_images: false
     type: openai_compat
   vision_service:
@@ -45,6 +45,11 @@ backends:
     model: __auto__
     supports_images: false
     type: openai_compat
+    cf_orch:
+      service: vllm
+      model_candidates:
+      - Qwen2.5-3B-Instruct
+      ttl_s: 300
   vllm_research:
     api_key: ''
     base_url: http://host.docker.internal:8000/v1

docker/web/Dockerfile

@@ -4,6 +4,8 @@ WORKDIR /app
 COPY web/package*.json ./
 RUN npm ci --prefer-offline
 COPY web/ ./
+ARG VITE_BASE_PATH=/
+ENV VITE_BASE_PATH=${VITE_BASE_PATH}
 RUN npm run build
 
 # Stage 2: serve

docker/web/nginx.conf

@@ -5,9 +5,13 @@ server {
     root /usr/share/nginx/html;
     index index.html;
 
-    # SPA fallback
+    # Proxy API calls to the FastAPI backend service
-    location / {
+    location /api/ {
-        try_files $uri $uri/ /index.html;
+        proxy_pass http://api:8601;
+        proxy_set_header Host $host;
+        proxy_set_header X-Real-IP $remote_addr;
+        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+        proxy_read_timeout 120s;
     }
 
     # Cache static assets
@@ -15,4 +19,9 @@ server {
         expires 1y;
         add_header Cache-Control "public, immutable";
     }
+
+    # SPA fallback — must come after API and assets
+    location / {
+        try_files $uri $uri/ /index.html;
+    }
 }

scripts/generate_cover_letter.py

@@ -26,13 +26,14 @@ LETTERS_DIR = _profile.docs_dir if _profile else Path.home() / "Documents" / "Jo
 LETTER_GLOB = "*Cover Letter*.md"
 
 # Background injected into every prompt so the model has the candidate's facts
-def _build_system_context() -> str:
+def _build_system_context(profile=None) -> str:
-    if not _profile:
+    p = profile or _profile
+    if not p:
         return "You are a professional cover letter writer. Write in first person."
-    parts = [f"You are writing cover letters for {_profile.name}. {_profile.career_summary}"]
+    parts = [f"You are writing cover letters for {p.name}. {p.career_summary}"]
-    if _profile.candidate_voice:
+    if p.candidate_voice:
         parts.append(
-            f"Voice and personality: {_profile.candidate_voice} "
+            f"Voice and personality: {p.candidate_voice} "
             "Write in a way that reflects these authentic traits — not as a checklist, "
             "but as a natural expression of who this person is."
         )
@@ -125,15 +126,17 @@ _MISSION_DEFAULTS: dict[str, str] = {
 }
 
 
-def _build_mission_notes() -> dict[str, str]:
+def _build_mission_notes(profile=None, candidate_name: str | None = None) -> dict[str, str]:
     """Merge user's custom mission notes with generic defaults."""
-    prefs = _profile.mission_preferences if _profile else {}
+    p = profile or _profile
+    name = candidate_name or _candidate
+    prefs = p.mission_preferences if p else {}
     notes = {}
     for industry, default_note in _MISSION_DEFAULTS.items():
         custom = (prefs.get(industry) or "").strip()
         if custom:
             notes[industry] = (
-                f"Mission alignment — {_candidate} shared: \"{custom}\". "
+                f"Mission alignment — {name} shared: \"{custom}\". "
                 "Para 3 should warmly and specifically reflect this authentic connection."
             )
         else:
@@ -144,12 +147,15 @@ def _build_mission_notes() -> dict[str, str]:
 _MISSION_NOTES = _build_mission_notes()
 
 
-def detect_mission_alignment(company: str, description: str) -> str | None:
+def detect_mission_alignment(
+    company: str, description: str, mission_notes: dict | None = None
+) -> str | None:
     """Return a mission hint string if company/JD matches a preferred industry, else None."""
+    notes = mission_notes if mission_notes is not None else _MISSION_NOTES
     text = f"{company} {description}".lower()
     for industry, signals in _MISSION_SIGNALS.items():
         if any(sig in text for sig in signals):
-            return _MISSION_NOTES[industry]
+            return notes[industry]
     return None
 
 
@@ -190,10 +196,14 @@ def build_prompt(
     examples: list[dict],
     mission_hint: str | None = None,
     is_jobgether: bool = False,
+    system_context: str | None = None,
+    candidate_name: str | None = None,
 ) -> str:
-    parts = [SYSTEM_CONTEXT.strip(), ""]
+    ctx = system_context if system_context is not None else SYSTEM_CONTEXT
+    name = candidate_name or _candidate
+    parts = [ctx.strip(), ""]
     if examples:
-        parts.append(f"=== STYLE EXAMPLES ({_candidate}'s past letters) ===\n")
+        parts.append(f"=== STYLE EXAMPLES ({name}'s past letters) ===\n")
         for i, ex in enumerate(examples, 1):
             parts.append(f"--- Example {i} ({ex['company']}) ---")
             parts.append(ex["text"])
@@ -231,13 +241,14 @@ def build_prompt(
     return "\n".join(parts)
 
 
-def _trim_to_letter_end(text: str) -> str:
+def _trim_to_letter_end(text: str, profile=None) -> str:
     """Remove repetitive hallucinated content after the first complete sign-off.
 
     Fine-tuned models sometimes loop after completing the letter. This cuts at
     the first closing + candidate name so only the intended letter is saved.
     """
-    candidate_first = (_profile.name.split()[0] if _profile else "").strip()
+    p = profile or _profile
+    candidate_first = (p.name.split()[0] if p else "").strip()
     pattern = (
         r'(?:Warm regards|Sincerely|Best regards|Kind regards|Thank you)[,.]?\s*\n+\s*'
         + (re.escape(candidate_first) if candidate_first else r'\w+(?:\s+\w+)?')
@@ -257,6 +268,8 @@ def generate(
     feedback: str = "",
     is_jobgether: bool = False,
     _router=None,
+    config_path: "Path | None" = None,
+    user_yaml_path: "Path | None" = None,
 ) -> str:
     """Generate a cover letter and return it as a string.
 
@@ -264,15 +277,29 @@ def generate(
     and requested changes are appended to the prompt so the LLM revises rather
     than starting from scratch.
 
+    user_yaml_path overrides the module-level profile — required in cloud mode
+    so each user's name/voice/mission prefs are used instead of the global default.
+
     _router is an optional pre-built LLMRouter (used in tests to avoid real LLM calls).
     """
+    # Per-call profile override (cloud mode: each user has their own user.yaml)
+    if user_yaml_path and Path(user_yaml_path).exists():
+        _prof = UserProfile(Path(user_yaml_path))
+    else:
+        _prof = _profile
+
+    sys_ctx = _build_system_context(_prof)
+    mission_notes = _build_mission_notes(_prof, candidate_name=(_prof.name if _prof else None))
+    candidate_name = _prof.name if _prof else _candidate
+
     corpus = load_corpus()
     examples = find_similar_letters(description or f"{title} {company}", corpus)
-    mission_hint = detect_mission_alignment(company, description)
+    mission_hint = detect_mission_alignment(company, description, mission_notes=mission_notes)
     if mission_hint:
         print(f"[cover-letter] Mission alignment detected for {company}", file=sys.stderr)
     prompt = build_prompt(title, company, description, examples,
-                          mission_hint=mission_hint, is_jobgether=is_jobgether)
+                          mission_hint=mission_hint, is_jobgether=is_jobgether,
+                          system_context=sys_ctx, candidate_name=candidate_name)
 
     if previous_result:
         prompt += f"\n\n---\nPrevious draft:\n{previous_result}"
@@ -281,8 +308,9 @@ def generate(
 
     if _router is None:
         sys.path.insert(0, str(Path(__file__).parent.parent))
-        from scripts.llm_router import LLMRouter
+        from scripts.llm_router import LLMRouter, CONFIG_PATH
-        _router = LLMRouter()
+        resolved = config_path if (config_path and Path(config_path).exists()) else CONFIG_PATH
+        _router = LLMRouter(resolved)
 
     print(f"[cover-letter] Generating for: {title} @ {company}", file=sys.stderr)
     print(f"[cover-letter] Style examples: {[e['company'] for e in examples]}", file=sys.stderr)
@@ -292,7 +320,7 @@ def generate(
     # max_tokens=1200 caps generation at ~900 words — enough for any cover letter
     # and prevents fine-tuned models from looping into repetitive garbage output.
     result = _router.complete(prompt, max_tokens=1200)
-    return _trim_to_letter_end(result)
+    return _trim_to_letter_end(result, _prof)
 
 
 def main() -> None:

scripts/preflight.py

@@ -47,7 +47,7 @@ OVERRIDE_YML = ROOT / "compose.override.yml"
 _SERVICES: dict[str, tuple[str, int, str, bool, bool]] = {
     "streamlit":       ("streamlit_port",       8501,  "STREAMLIT_PORT",       True,  False),
     "searxng":         ("searxng_port",          8888,  "SEARXNG_PORT",         True,  True),
-    "vllm":            ("vllm_port",             8000,  "VLLM_PORT",            True,  True),
+    # vllm removed — now managed by cf-orch (host process), not a Docker service
     "vision":          ("vision_port",           8002,  "VISION_PORT",          True,  True),
     "ollama":          ("ollama_port",          11434,  "OLLAMA_PORT",          True,  True),
     "ollama_research": ("ollama_research_port", 11435,  "OLLAMA_RESEARCH_PORT", True,  True),
@@ -65,7 +65,6 @@ _LLM_BACKENDS: dict[str, list[tuple[str, str]]] = {
 _DOCKER_INTERNAL: dict[str, tuple[str, int]] = {
     "ollama":          ("ollama",          11434),
     "ollama_research": ("ollama_research", 11434),  # container-internal port is always 11434
-    "vllm":            ("vllm",             8000),
     "vision":          ("vision",           8002),
     "searxng":         ("searxng",          8080),  # searxng internal port differs from host port
 }

scripts/task_runner.py

@@ -179,6 +179,9 @@ def _run_task(db_path: Path, task_id: int, task_type: str, job_id: int,
             import json as _json
             p = _json.loads(params or "{}")
             from scripts.generate_cover_letter import generate
+            _cfg_dir = Path(db_path).parent / "config"
+            _user_llm_cfg = _cfg_dir / "llm.yaml"
+            _user_yaml = _cfg_dir / "user.yaml"
             result = generate(
                 job.get("title", ""),
                 job.get("company", ""),
@@ -186,6 +189,8 @@ def _run_task(db_path: Path, task_id: int, task_type: str, job_id: int,
                 previous_result=p.get("previous_result", ""),
                 feedback=p.get("feedback", ""),
                 is_jobgether=job.get("source") == "jobgether",
+                config_path=_user_llm_cfg,
+                user_yaml_path=_user_yaml,
             )
             update_cover_letter(db_path, job_id, result)
 

tests/test_cover_letter_refinement.py

@@ -80,7 +80,8 @@ class TestTaskRunnerCoverLetterParams:
         captured = {}
 
         def mock_generate(title, company, description="", previous_result="", feedback="",
-                          is_jobgether=False, _router=None):
+                          is_jobgether=False, _router=None, config_path=None,
+                          user_yaml_path=None):
             captured.update({
                 "title": title, "company": company,
                 "previous_result": previous_result, "feedback": feedback,

web/src/router/index.ts

@@ -3,7 +3,7 @@ import { useAppConfigStore } from '../stores/appConfig'
 import { settingsGuard } from './settingsGuard'
 
 export const router = createRouter({
-  history: createWebHistory(),
+  history: createWebHistory(import.meta.env.BASE_URL),
   routes: [
     { path: '/',           component: () => import('../views/HomeView.vue') },
     { path: '/review',     component: () => import('../views/JobReviewView.vue') },

web/vite.config.ts

@@ -3,6 +3,7 @@ import vue from '@vitejs/plugin-vue'
 import UnoCSS from 'unocss/vite'
 
 export default defineConfig({
+  base: process.env.VITE_BASE_PATH || '/',
   plugins: [vue(), UnoCSS()],
   server: {
     host: '0.0.0.0',