feat(ops): backend prêt prod + SPA statique (FastAPI+Vite) + CI + systemd/nginx #2

cyranoaladin · 2025-10-02T07:16:30Z

Objet

Rendre le backend opérationnel “prod-like” et servir le frontend buildé (Vite) depuis la même origine, + artefacts Ops.

Points clés

Backend FastAPI :
- Montage StaticFiles sur / si frontend/dist présent (même origine que l’API).
- Endpoint /healthz enrichi (headers X-LLM-Provider/Embeddings).
- Endpoints démo A01/A02/RAG et catalogue patterns.
LLM/Embeddings :
- backend/core/llm.py : fabrique OpenAI/Gemini sans kwargs non supportés par Gemini.
- backend/core/embed.py : embeddings OpenAI ou Gemini.
- backend/core/tools.py : Tavily avec garde si clé absente.
Ingestion RAG :
- backend/ingestion/build_faiss.py + vars RAG_DATA_DIR / RAG_INDEX_DIR.
Frontend (Vite/React TS) :
- UI démo (Gallery 17, Tool Use, RAG, Reflection, etc.).
- ProviderToggle (changement in-memory des providers).
CI GitHub Actions :
- Lint (ruff), build frontend, smoke local (/healthz, api/a01/run).
Ops artefacts :
- ops/systemd/all-agentic.service (uvicorn, same-origin).
- ops/nginx/all-agentic.conf (reverse proxy + cache assets).
- backend/ops/env_templates/.env*.example + scripts/copy_env_templates.sh.

Test local rapide

python -m venv .venv && . .venv/bin/activate
pip install -U pip && pip install -r backend/requirements.txt
cd frontend && npm ci && npm run build && cd ..
uvicorn backend.app:app --host 127.0.0.1 --port 8010
# → Ouvrir http://127.0.0.1:8010/

…oc note

…ing in prod_check

…nx templates and CI

…eep .env loading order

Copilot

Pull Request Overview

Mise en place d’un environnement “prod-like” : API FastAPI servant aussi le frontend Vite (même origine), ajout des endpoints de démonstration agentiques (Reflection, Tool Use, RAG, etc.), scripts d’ingestion RAG, configuration Ops (systemd, nginx) et CI (lint + build + smoke).

Ajout d’une application FastAPI avec nombreux endpoints agents + RAG + configuration dynamique des providers (in‑memory).
Intégration d’un frontend React/Vite servi statiquement par FastAPI si dist présent.
Ajout d’outils Ops (scripts run/dev/ingest, systemd service, nginx conf, CI) et catalogue de patterns.

Reviewed Changes

Copilot reviewed 70 out of 74 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
uvicorn.pid	Ajout d’un artefact PID (exécutable) commité.
scripts/prod_check.sh	Script de probes endpoints (healthz / agents).
scripts/ingest.sh	Script construction index FAISS RAG.
scripts/dev_run.sh	Script d’environnement dev + build conditionnel frontend.
scripts/copy_env_templates.sh	Copie de templates .env (dev/prod).
run.sh	Script start/stop/status uvicorn + build frontend.
ops/systemd/all-agentic.service	Service systemd uvicorn (prod).
ops/nginx/all-agentic.conf	Reverse proxy nginx + cache assets.
frontend/vite.config.ts	Configuration Vite + plugin React.
frontend/tsconfig.node.json	TS config (node/build tooling).
frontend/tsconfig.json	TS project references.
frontend/tsconfig.app.json	TS config app React.
frontend/src/types/patterns.ts	Types TS du catalogue patterns.
frontend/src/pages/gallery.css	Styles grille cartes.
frontend/src/pages/PatternsCatalog.tsx	Catalogue filtrable des patterns.
frontend/src/pages/PatternDetails.tsx	Détails pattern + lancement démo.
frontend/src/pages/Gallery17.tsx	Page “Gallery 17” (cards).
frontend/src/main.tsx	Entrée React root render.
frontend/src/index.css	Styles globaux layout.
frontend/src/components/ProviderToggle.tsx	Sélecteur dynamique providers LLM/embeddings.
frontend/src/api/patterns.ts	Fonctions fetch API patterns.
frontend/src/api.ts	Client API (health / a01 / rag / a02).
frontend/src/agents/A09ToT.tsx	UI démo Tree-of-Thoughts.
frontend/src/agents/A08GraphMemory.tsx	UI démo Graph Memory.
frontend/src/agents/A07EpisodicSemantic.tsx	UI démo mémoire épisodique+sémantique.
frontend/src/agents/A06Blackboard.tsx	UI démo blackboard.
frontend/src/agents/A05PEV.tsx	UI démo PEV.
frontend/src/agents/A04Planning.tsx	UI démo planning.
frontend/src/App.tsx	Shell principal + navigation + pages.
frontend/src/App.css	Styles résiduels template Vite (partiellement inutilisés).
frontend/package.json	Dépendances frontend React/Vite.
frontend/index.html	Entrée HTML Vite.
frontend/eslint.config.js	Config ESLint TS/React.
frontend/README.md	README template Vite/React.
frontend/.gitignore	Ignore node_modules/dist etc.
backend/requirements.txt	Dépendances FastAPI, LangChain, embeddings, etc.
backend/ops/env_templates/.env.production.example	Template env prod.
backend/ops/env_templates/.env.example	Template env dev.
backend/ingestion/build_faiss.py	Construction index FAISS (RAG).
backend/ingestion/init.py	Init vide.
backend/core/tools.py	Wrapper outil Tavily avec garde clé.
backend/core/schemas.py	Pydantic schemas (agents, RAG, patterns).
backend/core/rag.py	Chargement index FAISS + retriever.
backend/core/patterns_data.py	Données catalogue patterns (17).
backend/core/llm.py	Fabrique LLM OpenAI/Gemini.
backend/core/graph_utils.py	Utilitaire passthrough.
backend/core/embed.py	Fabrique embeddings OpenAI/Gemini.
backend/core/init.py	Init vide.
backend/app.py	Application FastAPI + endpoints SPA/static.
backend/agents/a17_reflexive_metacognitive.py	Agent réflexif métacognitif.
backend/agents/a16_simulator_mitl.py	Agent simulateur MITL.
backend/agents/a15_rlhf_loop.py	Boucle RLHF-like.
backend/agents/a14_dry_run.py	Dry-run multi prompts reflection.
backend/agents/a13_chain_of_verification.py	Chain-of-Verification agent.
backend/agents/a12_self_ask.py	Self-Ask agent.
backend/agents/a11_meta_controller.py	Meta-controller router.
backend/agents/a10_ensemble_decision.py	Ensemble/vote agent.
backend/agents/a09_tree_of_thoughts.py	ToT agent (exploration).
backend/agents/a08_graph_memory.py	Extraction graphe simple.
backend/agents/a07_episodic_semantic.py	Mémoire épisodique+sémantique.
backend/agents/a06_blackboard.py	Blackboard agent.
backend/agents/a05_pev.py	Plan-Execute-Verify agent.
backend/agents/a04_planning.py	Planning agent.
backend/agents/a03_react.py	ReAct agent.
backend/agents/a02_tool_use.py	Tool Use agent.
backend/agents/a01_reflection.py	Reflection agent.
backend/agents/init.py	Init vide.
backend/init.py	Init vide.
DEPLOY_VPS.md	Instructions déploiement env prod.
.github/workflows/ci.yml	Workflow CI lint + build + smoke.

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

Copilot · 2025-10-02T07:20:10Z

scripts/prod_check.sh

+curl -sS "http://$HOST:$PORT/api/a01/run" \
+  -H 'Content-Type: application/json' \
+  -d '{"prompt":"Explique la loi des grands nombres simplement."}' \
+| python - <<'PY'
+import sys, json
+print(json.dumps(json.load(sys.stdin), indent=2, ensure_ascii=False))
+PY


The JSON pretty-print is only executed when curl fails because '||' is used instead of piping; as written, successful responses are not formatted. Replace '|| python - <<'PY'' with a pipe '|' so the curl output (on success) is passed to python for formatting.

Copilot · 2025-10-02T07:20:10Z

scripts/prod_check.sh

+curl -sS "http://$HOST:$PORT/api/rag/qa" \
+  -H 'Content-Type: application/json' \
+  -d '{"question":"Quels thèmes sont évoqués dans la démo RAG ?","k":5}' \
+| python - <<'PY'
+import sys, json
+print(json.dumps(json.load(sys.stdin), indent=2, ensure_ascii=False))
+PY


Same issue: using '||' causes the pretty-printer to run only on error; use a pipe '|' to format successful JSON output.

Copilot · 2025-10-02T07:20:11Z

scripts/prod_check.sh

+curl -sS "http://$HOST:$PORT/api/a02/run" \
+  -H 'Content-Type: application/json' \
+  -d '{"question":"Quelles ont été les annonces majeures de la dernière WWDC ?"}' \
+| python - <<'PY'
+import sys, json
+raw = sys.stdin.read().strip()
+if not raw:
+    print("[WARN] vide (peut-être Tavily absent)"); sys.exit(0)
+try:
+    print(json.dumps(json.loads(raw), indent=2, ensure_ascii=False))
+except Exception:
+    print("[WARN] réponse non JSON:")
+    print(raw)


The fallback pretty-print block executes only if curl fails due to '||'; to always inspect and format output, switch to a pipe '|' before python and handle non-JSON inside the script as already implemented.

Copilot · 2025-10-02T07:20:11Z

scripts/dev_run.sh

+
+# build frontend si nécessaire
+if [ -d frontend ] && [ ! -d frontend/dist ]; then
+  ( cd frontend && npm i && npm run build )


[nitpick] Use 'npm ci' instead of 'npm i' for deterministic builds in CI/dev scripts (it uses package-lock and fails on mismatch), improving reproducibility. Suggested change: '( cd frontend && npm ci && npm run build )'.

Suggested change

( cd frontend && npm i && npm run build )

( cd frontend && npm ci && npm run build )

Copilot · 2025-10-02T07:20:11Z

frontend/src/components/ProviderToggle.tsx

+      await fetch("/api/config/provider", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(cfg),
+      }).then((r) => r.json());
+      setMsg("✅ Appliqué. (In-memory) — Les prochains appels utiliseront ces valeurs.");


The response status is ignored; if the server returns a non-2xx status (e.g. 400/500 with HTML or error JSON), '.json()' may throw producing an unhelpful message. Check r.ok and throw a descriptive error before parsing JSON. Example: 'const r = await fetch(...); if(!r.ok) throw new Error(Failed: ${r.status}); const data = await r.json();'.

Copilot · 2025-10-02T07:20:12Z

backend/agents/a09_tree_of_thoughts.py

+    task: str
+    breadth: int
+    depth: int


The 'depth' parameter is never used to limit expansion; the graph always has an unconditional self-edge 'expand' -> 'expand' plus a direct edge to 'final', so depth control is ineffective and may loop more times than intended. Introduce a depth counter in state (e.g. increment on each expansion) and conditionally route to 'final' when the configured depth is reached instead of unconditional edges.

Copilot · 2025-10-02T07:20:12Z

backend/agents/a09_tree_of_thoughts.py

+    g = StateGraph(S)
+    g.add_node("init", n_init)
+    g.add_node("expand", n_expand_layer)
+    g.add_node("final", n_last_answer)
+    g.set_entry_point("init")
+    g.add_edge("init","expand")
+    g.add_edge("expand","expand")
+    g.add_edge("expand","final")


The 'depth' parameter is never used to limit expansion; the graph always has an unconditional self-edge 'expand' -> 'expand' plus a direct edge to 'final', so depth control is ineffective and may loop more times than intended. Introduce a depth counter in state (e.g. increment on each expansion) and conditionally route to 'final' when the configured depth is reached instead of unconditional edges.

Copilot · 2025-10-02T07:20:13Z

backend/app.py

+    b = int(body.get("breadth", 3))
+    d = int(body.get("depth", 2))


'breadth' and 'depth' are taken directly from user input without validation; large values could trigger excessive LLM calls and resource usage. Clamp to safe maxima (e.g. breadth<=5, depth<=5) or return 422 for out-of-range values before running the graph.

Suggested change

b = int(body.get("breadth", 3))

d = int(body.get("depth", 2))

try:

b = int(body.get("breadth", 3))

d = int(body.get("depth", 2))

except (ValueError, TypeError):

raise HTTPException(status_code=422, detail="breadth and depth must be integers")

if not (1 <= b <= 5) or not (1 <= d <= 5):

raise HTTPException(status_code=422, detail="breadth and depth must be between 1 and 5")

Copilot · 2025-10-02T07:20:13Z

backend/app.py

+@app.post("/api/config/provider")
+def set_provider_config(
+    llm: dict = Body(default={}),
+    embeddings: dict = Body(default={})
+):
+    """Met à jour os.environ pour les appels suivants (ne modifie pas .env sur disque)."""
+    if "provider" in llm:
+        os.environ["DEFAULT_LLM_PROVIDER"] = str(llm["provider"]).lower()
+    if "model" in llm:
+        os.environ["DEFAULT_LLM_MODEL"] = str(llm["model"])
+    if "provider" in embeddings:
+        os.environ["DEFAULT_EMBEDDING_PROVIDER"] = str(embeddings["provider"]).lower()
+    if "model" in embeddings:
+        os.environ["DEFAULT_EMBEDDING_MODEL"] = str(embeddings["model"])
+    return {"ok": True, "applied": get_provider_config()}


This endpoint mutates process-wide environment variables which only affect the current worker; in a multi-worker uvicorn deployment (workers>1) requests may see inconsistent provider settings. Consider keeping an in-memory configuration object (thread-safe) and documenting non-persistence, or providing a persistent config mechanism applied at startup across all workers.

Copilot · 2025-10-02T07:20:13Z

uvicorn.pid

@@ -0,0 +1 @@
+923988


Committing a runtime PID file can cause stale process management and unnecessary merge noise; it should be added to .gitignore and removed from version control.

Suggested change

923988

…un, restart) + .nvmrc; fix lint & port collisions - Add .nvmrc with Node v20.19.0 for consistent Node version - Create scripts/ensure_venv.sh: creates/validates venv, upgrades pip, installs deps - Create scripts/port_free.sh: frees port by killing uvicorn processes from this repo - Create scripts/dev_run.sh: complete dev environment (venv + front build + backend launch) - Create scripts/restart.sh: restarts dev environment cleanly - All scripts are POSIX-compliant with set -euo pipefail - Scripts handle port conflicts automatically (fallback to 8011) - Scripts are idempotent and can be run multiple times safely - Fix frontend build issues with clean node_modules reinstall - All scripts are executable and tested working

cyranoaladin and others added 4 commits October 2, 2025 08:30

chore(env): add backend/ops/env_templates templates + copy script + d…

dba6de2

…oc note

fix(ops): here-doc quoting & runtime ROOT detection; strict JSON quot…

10b805d

…ing in prod_check

ops: fix Gemini stack + Tavily guard; add middleware; add systemd/ngi…

1e47dbe

…nx templates and CI

style(ruff): fix E741/E702/E401/E402/F401/F821/F841 across backend; k…

ffcca2d

…eep .env loading order

Copilot AI review requested due to automatic review settings October 2, 2025 07:16

Copilot AI reviewed Oct 2, 2025

View reviewed changes

13bMetaWealth added 2 commits October 2, 2025 11:42

ci: fix ModuleNotFoundError by setting PYTHONPATH and using -m

fb8f17f

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(ops): backend prêt prod + SPA statique (FastAPI+Vite) + CI + systemd/nginx #2

feat(ops): backend prêt prod + SPA statique (FastAPI+Vite) + CI + systemd/nginx #2

Uh oh!

cyranoaladin commented Oct 2, 2025

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Copilot AI Oct 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	( cd frontend && npm i && npm run build )
	( cd frontend && npm ci && npm run build )

		b = int(body.get("breadth", 3))
		d = int(body.get("depth", 2))

feat(ops): backend prêt prod + SPA statique (FastAPI+Vite) + CI + systemd/nginx #2

Are you sure you want to change the base?

feat(ops): backend prêt prod + SPA statique (FastAPI+Vite) + CI + systemd/nginx #2

Uh oh!

Conversation

cyranoaladin commented Oct 2, 2025

Objet

Points clés

Test local rapide

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 2, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants