GustavoNogueira/A-Team-Security-Infra-Agent-Engineer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: terraform prompt generator, provider parameter matrix, and model penalties

Browse Source 
- Add Terraform Prompt Generator: AI-powered structured prompt generation
  with dedicated chat UI, curated model selection (7 models), conversation
  history with persistent sessions, and OCI TF resource reference context
- Provider parameter matrix: segment API params per model/provider,
  explicitly null unsupported params (freq/pres penalty, temperature)
  to prevent OCI SDK serialization errors
- Add penalties flag to model catalog: only GPT-4.1/4.1-mini/4o support
  frequency_penalty and presence_penalty
- Flat workspace enforcement: prohibit module blocks and duplicate
  variable declarations in Terraform system prompt
- Prompt Generator history: sidebar panel with session persistence,
  restore, rename/delete (agent_type tf-prompt)
This commit is contained in:
nogueiraguh
2026-03-11 13:55:59 -03:00
parent ab38e93516
commit 58d430c904
3 changed files with 355 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

239
backend/app.py
View File

@@ -62,38 +62,38 @@ security = HTTPBearer()
# _call_genai resolves the OCID for the configured region at runtime.
GENAI_MODELS = {
# ── Meta ──
"meta.llama-4-maverick-17b-128e-instruct-fp8": {"provider":"meta","name":"Meta Llama 4 Maverick","api_format":"GENERIC",
"meta.llama-4-maverick-17b-128e-instruct-fp8": {"provider":"meta","name":"Meta Llama 4 Maverick","api_format":"GENERIC","max_tokens":8192,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyah6tjdejjashngznsylutuhhvufukzb2g2ls54g2flsfq"}},
"meta.llama-4-scout-17b-16e-instruct": {"provider":"meta","name":"Meta Llama 4 Scout","api_format":"GENERIC",
"meta.llama-4-scout-17b-16e-instruct": {"provider":"meta","name":"Meta Llama 4 Scout","api_format":"GENERIC","max_tokens":8192,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyaw23hfc7mtvv5wef3gwvvaguyzqmhb5lx4r5s3y2xzc4a"}},
# ── Google ──
"google.gemini-2.5-pro": {"provider":"google","name":"Google Gemini 2.5 Pro","api_format":"GENERIC",
"google.gemini-2.5-pro": {"provider":"google","name":"Google Gemini 2.5 Pro","api_format":"GENERIC","max_tokens":65536,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyargceyuaysrjzo2metq2rinavayxqmpu7tkm6mmfojcvq"}},
"google.gemini-2.5-flash": {"provider":"google","name":"Google Gemini 2.5 Flash","api_format":"GENERIC",
"google.gemini-2.5-flash": {"provider":"google","name":"Google Gemini 2.5 Flash","api_format":"GENERIC","max_tokens":8192,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyaeo4ehrn25guuats5s45hnvswlhxo6riop275l2bkr2vq"}},
# ── OpenAI ──
"openai.gpt-5.2": {"provider":"openai","name":"OpenAI GPT-5.2","api_format":"GENERIC",
"openai.gpt-5.2": {"provider":"openai","name":"OpenAI GPT-5.2","api_format":"GENERIC","max_tokens":32768,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceya4fw3p5fddnexbfcurnz7spgkqb4mq4a6y5ubyv7777sa"}},
"openai.gpt-5.1": {"provider":"openai","name":"OpenAI GPT-5.1","api_format":"GENERIC",
"openai.gpt-5.1": {"provider":"openai","name":"OpenAI GPT-5.1","api_format":"GENERIC","max_tokens":32768,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceya3darth2ozqcfssb2bats5jitpgigllccajasdyqljnkq"}},
"openai.gpt-5-mini": {"provider":"openai","name":"OpenAI GPT-5 Mini","api_format":"GENERIC",
"openai.gpt-5-mini": {"provider":"openai","name":"OpenAI GPT-5 Mini","api_format":"GENERIC","max_tokens":16384,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceya3eain4n6v3edm4ryjvze5hnjouujd4vralxntfalwjaq"}},
"openai.gpt-4.1": {"provider":"openai","name":"OpenAI GPT-4.1 (Padrão)","api_format":"GENERIC",
"openai.gpt-4.1": {"provider":"openai","name":"OpenAI GPT-4.1 (Padrão)","api_format":"GENERIC","max_tokens":32768,"penalties":True,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyaa6g75r2qqmtzjaooqtlxv4lxkpcqp2jdd6plpq7yq7ea"}},
"openai.gpt-4.1-mini": {"provider":"openai","name":"OpenAI GPT-4.1 Mini","api_format":"GENERIC",
"openai.gpt-4.1-mini": {"provider":"openai","name":"OpenAI GPT-4.1 Mini","api_format":"GENERIC","max_tokens":16384,"penalties":True,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceya6pk3sxishpiexm2rb5sf4ytb5tsbz4to2g3g23smidaa"}},
"openai.gpt-4o": {"provider":"openai","name":"OpenAI GPT-4o","api_format":"GENERIC",
"openai.gpt-4o": {"provider":"openai","name":"OpenAI GPT-4o","api_format":"GENERIC","max_tokens":16384,"penalties":True,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyah7slrtboxdbfdy5cdspsfts62yumoclpdgwydopse7za"}},
"openai.o4-mini": {"provider":"openai","name":"OpenAI o4-mini","api_format":"GENERIC",
"openai.o4-mini": {"provider":"openai","name":"OpenAI o4-mini","api_format":"GENERIC","reasoning":True,"max_tokens":100000,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceya5ivfqp4fxlajoeg2cahlqtuuswagiv6a7dggpigy23bq"}},
"openai.o3": {"provider":"openai","name":"OpenAI o3","api_format":"GENERIC",
"openai.o3": {"provider":"openai","name":"OpenAI o3","api_format":"GENERIC","reasoning":True,"max_tokens":100000,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyalgnrukpjk6wm5zsf4jzkoneahgswhrk7kukkoagwnzma"}},
# ── xAI ──
"xai.grok-4": {"provider":"xai","name":"xAI Grok 4","api_format":"GENERIC",
"xai.grok-4": {"provider":"xai","name":"xAI Grok 4","api_format":"GENERIC","max_tokens":131072,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyaldmhg25is4nouena4oa2pj4nvwgfeempo4syiaazukia"}},
"xai.grok-3": {"provider":"xai","name":"xAI Grok 3","api_format":"GENERIC",
"xai.grok-3": {"provider":"xai","name":"xAI Grok 3","api_format":"GENERIC","max_tokens":131072,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyag3w2xk76vlahjujj2gdfeuzhflt25gbo3bxidlsqfjla"}},
"xai.grok-3-mini-fast": {"provider":"xai","name":"xAI Grok 3 Mini Fast","api_format":"GENERIC",
"xai.grok-3-mini-fast": {"provider":"xai","name":"xAI Grok 3 Mini Fast","api_format":"GENERIC","max_tokens":131072,
"ocids":{"us-ashburn-1":"ocid1.generativeaimodel.oc1.iad.amaaaaaask7dceyauvjoll2repj5pbtkk7pinwj57ex3lkehzpxd6v6rxscq"}},
}
@@ -197,6 +197,7 @@ def init_db():
top_k INTEGER DEFAULT 1,
frequency_penalty REAL DEFAULT 0,
presence_penalty REAL DEFAULT 0,
reasoning_effort TEXT,
is_default INTEGER DEFAULT 0,
system_prompt TEXT DEFAULT '',
created_at TEXT DEFAULT (datetime('now')),
@@ -353,7 +354,7 @@ def init_db():
c.execute("DELETE FROM config_logs WHERE created_at < datetime('now', '-30 days')")
c.execute("DELETE FROM chat_logs WHERE created_at < datetime('now', '-30 days')")
# ── Migrations ──
for col in ["system_prompt TEXT DEFAULT ''"]:
for col in ["system_prompt TEXT DEFAULT ''", "reasoning_effort TEXT"]:
try:
c.execute(f"ALTER TABLE genai_configs ADD COLUMN {col}")
except sqlite3.OperationalError:
@@ -518,6 +519,7 @@ class ChatMsg(BaseModel):
temperature: Optional[float] = None; max_tokens: Optional[int] = None
top_p: Optional[float] = None; top_k: Optional[int] = None
frequency_penalty: Optional[float] = None; presence_penalty: Optional[float] = None
reasoning_effort: Optional[str] = None
use_tools: Optional[bool] = True
class RunReportReq(BaseModel):
config_id: str; regions: Optional[List[str]] = None
@@ -530,6 +532,7 @@ class GenAIConfigReq(BaseModel):
serving_type: str = "ON_DEMAND"; dedicated_endpoint_id: Optional[str] = None
temperature: float = 1; max_tokens: int = 6000; top_p: float = 0.95
top_k: int = 1; frequency_penalty: float = 0; presence_penalty: float = 0
reasoning_effort: Optional[str] = None
is_default: bool = False
class IngestDocReq(BaseModel):
adb_config_id: str; documents: List[Dict[str, Any]]; table_name: Optional[str] = None
@@ -1638,11 +1641,11 @@ async def save_genai(req: GenAIConfigReq, u=Depends(require("admin","user"))):
c.execute(
"""INSERT INTO genai_configs (id,user_id,name,oci_config_id,model_id,model_ocid,compartment_id,
genai_region,endpoint,serving_type,dedicated_endpoint_id,temperature,max_tokens,top_p,top_k,
frequency_penalty,presence_penalty,is_default) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
frequency_penalty,presence_penalty,reasoning_effort,is_default) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)""",
(gid, u["id"], req.name, req.oci_config_id, req.model_id, req.model_ocid,
req.compartment_id, req.genai_region, ep, req.serving_type, req.dedicated_endpoint_id,
req.temperature, req.max_tokens, req.top_p, req.top_k,
req.frequency_penalty, req.presence_penalty, int(req.is_default)))
req.frequency_penalty, req.presence_penalty, req.reasoning_effort, int(req.is_default)))
_audit(u["id"], u["username"], "save_genai_config", gid, req.model_id)
_config_log("genai", gid, req.name, "success", "save", f"Modelo salvo: {req.model_id} ({req.genai_region})", u["id"], u["username"])
return {"id": gid, "model_id": req.model_id, "endpoint": ep}
@@ -1672,11 +1675,11 @@ async def update_genai(gid: str, req: GenAIConfigReq, u=Depends(require("admin",
c.execute(
"""UPDATE genai_configs SET name=?,oci_config_id=?,model_id=?,model_ocid=?,compartment_id=?,
genai_region=?,endpoint=?,serving_type=?,dedicated_endpoint_id=?,temperature=?,max_tokens=?,
top_p=?,top_k=?,frequency_penalty=?,presence_penalty=?,is_default=? WHERE id=?""",
top_p=?,top_k=?,frequency_penalty=?,presence_penalty=?,reasoning_effort=?,is_default=? WHERE id=?""",
(req.name, req.oci_config_id, req.model_id, req.model_ocid,
req.compartment_id, req.genai_region, ep, req.serving_type, req.dedicated_endpoint_id,
req.temperature, req.max_tokens, req.top_p, req.top_k,
req.frequency_penalty, req.presence_penalty, int(req.is_default), gid))
req.frequency_penalty, req.presence_penalty, req.reasoning_effort, int(req.is_default), gid))
_audit(u["id"], u["username"], "update_genai_config", gid, req.model_id)
_config_log("genai", gid, req.name, "success", "save", f"Modelo atualizado: {req.model_id} ({req.genai_region})", u["id"], u["username"])
return {"id": gid, "model_id": req.model_id, "endpoint": ep}
@@ -1846,7 +1849,8 @@ def _call_genai(gc: dict, message: str, history: list = None, tools: list = None
if system_prompt:
chat_request.preamble_override = system_prompt
chat_request.message = message
chat_request.max_tokens = int(gc.get("max_tokens", 6000))
cohere_max = model_info.get("max_tokens", 4096)
chat_request.max_tokens = min(int(gc.get("max_tokens", 4096)), cohere_max)
chat_request.temperature = float(gc.get("temperature", 1))
chat_request.frequency_penalty = float(gc.get("frequency_penalty", 0))
chat_request.presence_penalty = float(gc.get("presence_penalty", 0))
@@ -1887,20 +1891,54 @@ def _call_genai(gc: dict, message: str, history: list = None, tools: list = None
chat_request.api_format = oci.generative_ai_inference.models.BaseChatRequest.API_FORMAT_GENERIC
provider = model_info.get("provider", "")
is_openai = provider == "openai"
is_xai = provider == "xai"
# OpenAI models use max_completion_tokens instead of max_tokens
# and do not support top_k
if is_openai:
chat_request.max_completion_tokens = int(gc.get("max_tokens", 6000))
is_reasoning = model_info.get("reasoning", False)
# Clamp max_tokens to model limit
model_max = model_info.get("max_tokens", 16384)
requested_tokens = int(gc.get("max_tokens", 6000))
clamped_tokens = min(requested_tokens, model_max)
# ── Parameter matrix per provider ──
# OpenAI reasoning (o3, o4-mini): max_completion_tokens + reasoning_effort only
if is_reasoning:
chat_request.max_completion_tokens = clamped_tokens
re = gc.get("reasoning_effort")
if re and hasattr(chat_request, "reasoning_effort"):
chat_request.reasoning_effort = re
# Explicitly unset unsupported params
chat_request.temperature = None
chat_request.top_p = None
chat_request.frequency_penalty = None
chat_request.presence_penalty = None
# OpenAI standard (GPT-4.1, GPT-5.x, GPT-4o): max_completion_tokens, temperature, top_p
# freq/pres penalty only for models with "penalties":True (GPT-4.1, GPT-4.1-mini, GPT-4o)
elif is_openai:
chat_request.max_completion_tokens = clamped_tokens
chat_request.temperature = float(gc.get("temperature", 1))
chat_request.top_p = float(gc.get("top_p", 0.95))
if model_info.get("penalties"):
chat_request.frequency_penalty = float(gc.get("frequency_penalty", 0))
chat_request.presence_penalty = float(gc.get("presence_penalty", 0))
else:
chat_request.frequency_penalty = None
chat_request.presence_penalty = None
# xAI Grok: max_tokens, temperature, top_p (no top_k, no freq/pres penalty)
elif provider == "xai":
chat_request.max_tokens = clamped_tokens
chat_request.temperature = float(gc.get("temperature", 1))
chat_request.top_p = float(gc.get("top_p", 0.95))
# Explicitly unset unsupported params to prevent SDK serialization
chat_request.frequency_penalty = None
chat_request.presence_penalty = None
# Meta Llama: max_tokens, temperature, top_p, top_k (no freq/pres penalty)
# Google Gemini: max_tokens, temperature, top_p, top_k (no freq/pres penalty)
else:
chat_request.max_tokens = int(gc.get("max_tokens", 6000))
if not is_xai:
chat_request.top_k = int(gc.get("top_k", 1))
chat_request.temperature = float(gc.get("temperature", 1))
if not is_xai:
chat_request.frequency_penalty = float(gc.get("frequency_penalty", 0))
chat_request.presence_penalty = float(gc.get("presence_penalty", 0))
chat_request.top_p = float(gc.get("top_p", 0.95))
chat_request.max_tokens = clamped_tokens
chat_request.temperature = float(gc.get("temperature", 1))
chat_request.top_p = float(gc.get("top_p", 0.95))
chat_request.top_k = int(gc.get("top_k", 1))
# Explicitly unset unsupported params to prevent SDK serialization
chat_request.frequency_penalty = None
chat_request.presence_penalty = None
messages = []
if system_prompt:
@@ -2164,6 +2202,11 @@ Se pedirem AWS, Azure, GCP ou outro provider, recuse educadamente.
- Remote peering: APENAS UM LADO do par define `peer_id` e `peer_region_name`. O outro lado é criado SEM `peer_id`. Nunca gere dois RPCs apontando um para o outro (causa Cycle error). Exemplo: RPC_mad1 (sem peer_id) e RPC_mad3 (com peer_id = RPC_mad1.id, peer_region_name = var.region).
- Associação de route table com subnet deve ser via `route_table_id` na subnet ou `oci_core_route_table_attachment`.
- `oci_network_firewall_network_firewall` exige `network_firewall_policy_id`.
### Proibições absolutas — módulos e variáveis
- NUNCA use `module` blocks. Este workspace é flat (um único diretório). Não existem subdiretórios `modules/`. Toda a infraestrutura deve ser definida diretamente com `resource` e `data` blocks.
- NUNCA declare a mesma `variable` em mais de um arquivo. Todas as variáveis devem estar em `variables.tf` e SOMENTE lá. Os outros arquivos (.tf) apenas referenciam `var.nome`.
- Antes de gerar, verifique se uma variável já foi declarada. Se já existe em `variables.tf`, NÃO redeclare.
"""
# ── Terraform OCI Resource Reference (generated at build time, updatable at runtime) ──
@@ -3821,6 +3864,7 @@ def _chat_start(msg: ChatMsg, u, attachments: list = None, agent_type: str = "ch
if msg.top_k is not None: genai_cfg["top_k"] = msg.top_k
if msg.frequency_penalty is not None: genai_cfg["frequency_penalty"] = msg.frequency_penalty
if msg.presence_penalty is not None: genai_cfg["presence_penalty"] = msg.presence_penalty
if msg.reasoning_effort is not None: genai_cfg["reasoning_effort"] = msg.reasoning_effort
elif msg.model_id and msg.oci_config_id:
with db() as c:
oci_row = c.execute("SELECT * FROM oci_configs WHERE id=?", (msg.oci_config_id,)).fetchone()
@@ -3845,6 +3889,7 @@ def _chat_start(msg: ChatMsg, u, attachments: list = None, agent_type: str = "ch
"top_k": msg.top_k if msg.top_k is not None else 1,
"frequency_penalty": msg.frequency_penalty if msg.frequency_penalty is not None else 0.0,
"presence_penalty": msg.presence_penalty if msg.presence_penalty is not None else 0.0,
"reasoning_effort": msg.reasoning_effort,
}
if not genai_cfg:
@@ -3976,9 +4021,11 @@ async def _chat_background(mid: str, sid: str, msg: ChatMsg, user: dict, genai_c
base_prompt = cfg_dict.get("system_prompt", "")
cfg_dict["system_prompt"] = f"{base_prompt}\n\n{config_hint}" if base_prompt else config_hint
# ── Terraform agent: boost max_tokens for complex infra code ──
# ── Terraform agent: boost max_tokens to model limit ──
if agent_type == "terraform":
cfg_dict["max_tokens"] = max(int(cfg_dict.get("max_tokens", 6000)), 100000)
tf_model_info = GENAI_MODELS.get(cfg_dict.get("model_id", ""), {})
tf_model_max = tf_model_info.get("max_tokens", 32768)
cfg_dict["max_tokens"] = tf_model_max
# ── Inject existing OCI resources for terraform agent ──
if agent_type == "terraform" and active_oci_id:
@@ -4184,6 +4231,7 @@ async def chat_with_files(
top_k: Optional[int] = Form(None),
frequency_penalty: Optional[float] = Form(None),
presence_penalty: Optional[float] = Form(None),
reasoning_effort: Optional[str] = Form(None),
files: List[UploadFile] = File(default=[]),
u=Depends(current_user),
):
@@ -4226,6 +4274,7 @@ async def chat_with_files(
top_k=top_k,
frequency_penalty=frequency_penalty,
presence_penalty=presence_penalty,
reasoning_effort=reasoning_effort,
)
sid, mid_or_result, genai_cfg = _chat_start(msg, u, attachments=attachments if attachments else None)
@@ -4510,6 +4559,120 @@ async def tf_list_resources(oci_config_id: str = Query(...), compartment_id: str
raise HTTPException(500, str(e)[:500])
TFP_SYSTEM_PROMPT = """Você é um arquiteto sênior de infraestrutura OCI com profundo conhecimento do Terraform OCI Provider.
O usuário vai descrever a infraestrutura desejada em linguagem natural. Sua tarefa é gerar um **prompt detalhado e estruturado** que será enviado a um agente Terraform para criar o código HCL.
### Formato do Prompt Gerado
Retorne APENAS o prompt estruturado (não gere código Terraform). Use markdown:
1. **Título**: Uma linha descrevendo o projeto (ex: "## Infraestrutura Multi-Region HA para Aplicação Web")
2. **Arquitetura**: Diagrama textual ou descrição da topologia
3. **Recursos por Categoria**: Organizados em seções ## (Networking, Compute, Storage, Database, Security, Observability)
- Para cada recurso: nome, especificações técnicas, relacionamentos
4. **Networking**: Sempre inclua CIDRs (10.0.0.0/16 para VCN, /24 para subnets), gateways necessários, route tables, security lists/NSGs
5. **Dependências**: Liste dependências entre recursos (ex: "Compute depende de Subnet, que depende de VCN")
6. **Requisitos Técnicos**: Seção final com regras para o agente Terraform:
- Separação em arquivos (provider.tf, variables.tf, vcn.tf, subnets.tf, compute.tf, outputs.tf)
- Variáveis obrigatórias (compartment_id, region, ssh_public_key, etc.)
- Naming convention (ex: proj-env-resource)
- Outputs necessários (IPs, OCIDs, endpoints)
- Tags padrão (Environment, Project, ManagedBy=Terraform)
### Regras de Inferência
- Se pedir VCN, inclua automaticamente: subnets (pública + privada), internet gateway, NAT gateway, service gateway, route tables, security lists
- Se pedir Compute, inclua: boot volume, NSG, cloud-init básico, shape e imagem
- Se pedir Database, inclua: subnet privada dedicada, NSG com porta do DB, backup policy
- Se pedir OKE, inclua: VCN com topologia para K8s (API endpoint subnet, workers subnet, pods subnet, services LB subnet)
- Se mencionar multi-region, use providers aliased e detalhe RPC/DRG
- Se mencionar HA/DR, inclua redundância cross-AD ou cross-region
- Se mencionar segurança/CIS, inclua Vault, Cloud Guard, WAF, logging, encryption at rest
### Restrições do OCI Provider
- RPC (Remote Peering Connection): NÃO usar oci_core_drg_attachment para RPC. O attachment é criado automaticamente pelo recurso oci_core_remote_peering_connection
- Cada VCN suporta apenas 1 DRG attachment — não criar duplicados
- DRG attachment type para VCN é "VCN", não "REMOTE_PEERING_CONNECTION"
### Restrições do Workspace
- NUNCA sugira o uso de `module` blocks. O workspace é flat (diretório único, sem subdiretórios). Toda infraestrutura deve ser definida com `resource` e `data` blocks diretamente.
- Variáveis devem ser declaradas SOMENTE em `variables.tf`. Nunca duplicar declarações de variáveis entre arquivos.
### Tom
- Seja técnico e preciso
- Use português brasileiro
- Não gere código, apenas o prompt estruturado"""
class TfPromptReq(BaseModel):
message: str
genai_config: Optional[dict] = None
history: Optional[list] = None
session_id: Optional[str] = None
@app.post("/api/terraform/generate-prompt")
async def tf_generate_prompt(req: TfPromptReq, u=Depends(current_user)):
gc = None
if req.genai_config:
if req.genai_config.get("genai_config_id"):
with db() as c:
row = c.execute("SELECT * FROM genai_configs WHERE id=?", (req.genai_config["genai_config_id"],)).fetchone()
if row: gc = dict(row)
elif req.genai_config.get("oci_config_id"):
oci_id = req.genai_config["oci_config_id"]
with db() as c:
oc = c.execute("SELECT * FROM oci_configs WHERE id=?", (oci_id,)).fetchone()
if oc:
region = req.genai_config.get("genai_region") or oc["region"]
compartment = _safe_dec(dict(oc).get("compartment_id") or oc["tenancy_ocid"])
gc = {
"oci_config_id": oci_id, "model_id": req.genai_config.get("model_id", "openai.gpt-4.1"),
"model_ocid": None, "compartment_id": compartment, "genai_region": region,
"endpoint": f"https://inference.generativeai.{region}.oci.oraclecloud.com",
"serving_type": "ON_DEMAND", "dedicated_endpoint_id": None,
"temperature": 0.7, "max_tokens": 4000, "top_p": 0.9,
}
if not gc:
with db() as c:
row = c.execute("SELECT * FROM genai_configs WHERE is_default=1").fetchone()
if not row:
row = c.execute("SELECT * FROM genai_configs LIMIT 1").fetchone()
if row: gc = dict(row)
if not gc:
raise HTTPException(400, "Nenhuma configuração GenAI disponível")
# Session persistence
is_new = not req.session_id
sid = req.session_id or str(uuid.uuid4())
with db() as c:
c.execute("INSERT INTO chat_messages (id,session_id,user_id,role,content,model_id,status) VALUES (?,?,?,?,?,?,?)",
(str(uuid.uuid4()), sid, u["id"], "user", req.message, gc.get("model_id"), "done"))
if is_new:
title = (req.message or "Prompt Generator")[:80].strip()
c.execute("INSERT OR IGNORE INTO chat_sessions (id,user_id,agent_type,title) VALUES (?,?,?,?)",
(sid, u["id"], "tf-prompt", title))
else:
c.execute("UPDATE chat_sessions SET updated_at=datetime('now') WHERE id=?", (sid,))
# Load TF resource reference for context
tf_ref = _load_tf_resource_reference()
system_with_ref = TFP_SYSTEM_PROMPT
if tf_ref:
system_with_ref += f"\n\n### Referência de Recursos OCI Terraform Disponíveis\nUse esta referência para garantir que os recursos mencionados no prompt existam no provider OCI:\n\n{tf_ref}"
gc["system_prompt"] = system_with_ref
gc["temperature"] = 0.7
gc["max_tokens"] = 4000
# Build history from conversation (normalize roles to lowercase for _call_genai)
hist = None
if req.history:
hist = [{"role": "user" if m.get("role","").upper() in ("USER","user") else "assistant",
"content": m.get("content", "")} for m in req.history]
try:
answer, _, _ = _call_genai(gc, req.message, history=hist)
# Save assistant response
with db() as c:
c.execute("INSERT INTO chat_messages (id,session_id,user_id,role,content,model_id,status) VALUES (?,?,?,?,?,?,?)",
(str(uuid.uuid4()), sid, u["id"], "assistant", answer, gc.get("model_id"), "done"))
return {"prompt": answer, "session_id": sid}
except Exception as e:
log.error(f"tf_generate_prompt error: {e}")
raise HTTPException(500, str(e)[:500])
@app.post("/api/terraform/chat")
async def terraform_chat(msg: ChatMsg, bg: BackgroundTasks, u=Depends(current_user)):
sid, mid_or_result, genai_cfg = _chat_start(msg, u, agent_type="terraform")
@@ -5276,12 +5439,12 @@ async def import_config(file: UploadFile = File(...), u=Depends(require("admin")
for cfg in data.get("genai_configs", []):
if c.execute("SELECT 1 FROM genai_configs WHERE id=?", (cfg["id"],)).fetchone():
counts["skipped"] += 1; continue
c.execute("INSERT INTO genai_configs (id,user_id,name,oci_config_id,model_id,model_ocid,compartment_id,genai_region,endpoint,serving_type,dedicated_endpoint_id,temperature,max_tokens,top_p,top_k,frequency_penalty,presence_penalty,is_default,system_prompt,created_at) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)",
c.execute("INSERT INTO genai_configs (id,user_id,name,oci_config_id,model_id,model_ocid,compartment_id,genai_region,endpoint,serving_type,dedicated_endpoint_id,temperature,max_tokens,top_p,top_k,frequency_penalty,presence_penalty,reasoning_effort,is_default,system_prompt,created_at) VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)",
(cfg["id"], cfg.get("user_id", u["id"]), cfg.get("name",""), cfg.get("oci_config_id",""), cfg.get("model_id",""),
cfg.get("model_ocid"), cfg.get("compartment_id",""), cfg.get("genai_region",""), cfg.get("endpoint",""),
cfg.get("serving_type","ON_DEMAND"), cfg.get("dedicated_endpoint_id"), cfg.get("temperature",1),
cfg.get("max_tokens",6000), cfg.get("top_p",0.95), cfg.get("top_k",1), cfg.get("frequency_penalty",0),
cfg.get("presence_penalty",0), cfg.get("is_default",0), cfg.get("system_prompt",""), cfg.get("created_at")))
cfg.get("presence_penalty",0), cfg.get("reasoning_effort"), cfg.get("is_default",0), cfg.get("system_prompt",""), cfg.get("created_at")))
counts["genai_configs"] += 1
# MCP servers
for srv in data.get("mcp_servers", []):