first

services/asf-ai/app/__init__.py

@@ -0,0 +1,54 @@
+"""asf-ai: ASF AI Gateway — "l'ufficio analisti".
+
+*Business code NEVER calls a specific LLM provider directly.*
+It calls typed capability methods on this gateway.
+
+Provider adapters (Mock, OpenAI-compat, NVIDIA, local, customer-hosted, ...)
+implement the low-level interface. Routing, priority queues, tenant AI
+policy, usage metering and cost tracking live here.
+"""
+
+__version__ = "0.1.0"
+
+from .capabilities import (
+    CapabilityRequest,
+    CapabilityResponse,
+    ChatRequest,
+    ChatResponse,
+    CodeAnalysisRequest,
+    CodeAnalysisResult,
+    TestGenerationRequest,
+    GeneratedTests,
+    RequirementDecomposeRequest,
+    RequirementDecomposition,
+    TraceContext,
+)
+from .gateway import AIGateway, AIGatewayConfig
+from .providers import (
+    AIProvider,
+    MockProvider,
+    ProviderRegistry,
+    build_default_gateway,
+)
+from .usage import UsageRecord
+
+__all__ = [
+    "CapabilityRequest",
+    "CapabilityResponse",
+    "ChatRequest",
+    "ChatResponse",
+    "CodeAnalysisRequest",
+    "CodeAnalysisResult",
+    "TestGenerationRequest",
+    "GeneratedTests",
+    "RequirementDecomposeRequest",
+    "RequirementDecomposition",
+    "TraceContext",
+    "AIGateway",
+    "AIGatewayConfig",
+    "AIProvider",
+    "MockProvider",
+    "ProviderRegistry",
+    "build_default_gateway",
+    "UsageRecord",
+]

services/asf-ai/app/capabilities.py

@@ -0,0 +1,133 @@
+"""Typed capability requests / responses.
+
+Business services construct these Pydantic models and call methods on the
+AIGateway. They never know which concrete model or provider was chosen.
+"""
+
+from __future__ import annotations
+
+from typing import Any, Optional, Sequence
+from uuid import UUID
+
+from pydantic import BaseModel, Field
+
+
+class TraceContext(BaseModel):
+    """Context propagated through every call for audit / logging."""
+
+    tenant_id: UUID
+    project_id: Optional[UUID] = None
+    user_id: Optional[UUID] = None
+    job_id: Optional[UUID] = None
+    correlation_id: Optional[str] = None
+
+
+class CapabilityRequest(BaseModel):
+    """Base marker for every capability request supported by the gateway."""
+
+    trace: TraceContext
+    preferred_model: Optional[str] = Field(
+        None, max_length=128, description="Hint only. Gateway's routing decision is authoritative."
+    )
+    priority: str = "normal"
+
+
+class CapabilityResponse(BaseModel):
+    """Every response carries at least trace + usage."""
+
+    provider: str
+    model_used: str
+    input_tokens: Optional[int] = None
+    output_tokens: Optional[int] = None
+    duration_ms: Optional[int] = None
+    estimated_cost: Optional[float] = None
+    execution_id: UUID
+    raw: Optional[dict[str, Any]] = Field(
+        default=None, description="Raw provider output, preserved for debugging."
+    )
+
+
+# ============================================================================
+# Individual capabilities
+# ============================================================================
+
+
+class ChatMessage(BaseModel):
+    role: str = Field(..., pattern="^(system|user|assistant|tool)$")
+    content: str
+
+
+class ChatRequest(CapabilityRequest):
+    messages: Sequence[ChatMessage]
+    temperature: float = Field(0.2, ge=0.0, le=2.0)
+    max_tokens: Optional[int] = Field(None, ge=1)
+    context_summary: Optional[str] = Field(
+        None, description="Short plaintext summary of the Context Engine assembly."
+    )
+
+
+class ChatResponse(CapabilityResponse):
+    reply: str
+    references: list[str] = Field(default_factory=list)
+
+
+class CodeAnalysisRequest(CapabilityRequest):
+    file_path: Optional[str] = None
+    language: Optional[str] = None
+    code_snippet: str = Field(..., min_length=1)
+    analysis_types: Sequence[str] = Field(
+        default_factory=lambda: ["bugs", "complexity", "security", "style"],
+        description="Which kinds of analyses to run.",
+    )
+
+
+class Finding(BaseModel):
+    severity: str = Field(..., pattern="^(info|low|medium|high|critical)$")
+    kind: str
+    message: str
+    line_start: Optional[int] = None
+    line_end: Optional[int] = None
+    suggestion: Optional[str] = None
+
+
+class CodeAnalysisResult(CapabilityResponse):
+    summary: str
+    findings: list[Finding] = Field(default_factory=list)
+
+
+class TestGenerationRequest(CapabilityRequest):
+    file_path: Optional[str] = None
+    language: str = "python"
+    test_framework: str = "pytest"
+    code_snippet: str = Field(..., min_length=1)
+    goal: str = "Generate unit tests covering happy path + main failure modes"
+    existing_tests: Optional[str] = None
+
+
+class GeneratedTests(CapabilityResponse):
+    test_cases: list[str] = Field(default_factory=list)
+    fixtures_needed: list[str] = Field(default_factory=list)
+    coverage_notes: str = ""
+
+
+class RequirementDecomposeRequest(CapabilityRequest):
+    requirement_text: str = Field(..., min_length=1)
+    context: Optional[str] = None
+    existing_subtasks: Sequence[str] = Field(default_factory=list)
+    max_subtasks: int = Field(20, ge=1, le=100)
+
+
+class SubTask(BaseModel):
+    title: str
+    description: str
+    estimated_complexity: str = Field(..., pattern="^(xs|s|m|l|xl)$")
+    depends_on: list[int] = Field(
+        default_factory=list,
+        description="Index (0-based) of sibling subtasks this one depends on.",
+    )
+
+
+class RequirementDecomposition(CapabilityResponse):
+    subtasks: list[SubTask] = Field(default_factory=list)
+    architecture_notes: list[str] = Field(default_factory=list)
+    open_questions: list[str] = Field(default_factory=list)

services/asf-ai/app/config.py

@@ -0,0 +1,29 @@
+"""asf-ai settings surface (minimal; extensible)."""
+
+from __future__ import annotations
+
+from functools import lru_cache
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+
+class AISettings(BaseSettings):
+    mongo_uri: str = "mongodb://172.19.40.102:27017,172.19.40.103:27017/asf?replicaSet=rs0"
+    mongo_db: str = "asf"
+
+    default_provider: str = "mock"
+
+    # Provider-specific (optional — only required if the chosen provider is used)
+    openai_api_key: str | None = None
+    anthropic_api_key: str | None = None
+    nvidia_api_key: str | None = None
+
+    model_config = SettingsConfigDict(
+        env_file=".env",
+        env_file_encoding="utf-8",
+        extra="ignore",
+    )
+
+
+@lru_cache(maxsize=1)
+def get_settings() -> AISettings:
+    return AISettings()

services/asf-ai/app/gateway.py

@@ -0,0 +1,185 @@
+"""AIGateway — the high-level capability surface called by business code.
+
+It handles:
+  1. Routing: picks which provider + model to use given the request.
+  2. Policy enforcement: tenant-level AI policy (which providers/models are
+     allowed; budget caps will plug in here cleanly later).
+  3. Usage recording: emits UsageRecords for metering + future billing.
+  4. Error translation (provider-specific → stable ASF exceptions).
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from typing import Callable, Optional
+
+import structlog
+
+from .capabilities import (
+    ChatRequest,
+    ChatResponse,
+    CodeAnalysisRequest,
+    CodeAnalysisResult,
+    TestGenerationRequest,
+    GeneratedTests,
+    RequirementDecomposeRequest,
+    RequirementDecomposition,
+)
+from .providers import (
+    AIProvider,
+    ProviderRegistry,
+    default_registry,
+)
+from .usage import UsageRecord
+
+log = structlog.get_logger(__name__)
+
+
+@dataclass
+class AIGatewayConfig:
+    default_provider: str = "mock"
+    default_model_by_capability: dict[str, str] = field(default_factory=dict)
+    # Hook for future tenant policy layer: given (tenant_id, capability),
+    # return the provider name to route to. None → use default_provider.
+    policy_selector: Optional[Callable[..., Optional[str]]] = None
+
+
+class AIGateway:
+    """Entry point for all ASF AI operations."""
+
+    def __init__(
+        self,
+        *,
+        config: Optional[AIGatewayConfig] = None,
+        registry: Optional[ProviderRegistry] = None,
+    ) -> None:
+        self._config = config or AIGatewayConfig()
+        self._registry = registry or default_registry()
+        self._usage_callbacks: list[Callable[[UsageRecord], None]] = []
+
+    # --- callback hooks (future usage metering / billing) ---
+    def on_usage(self, cb: Callable[[UsageRecord], None]) -> None:
+        self._usage_callbacks.append(cb)
+
+    # --- routing helper ---
+    def _route(self, req, capability: str) -> AIProvider:
+        tenant_id = str(req.trace.tenant_id) if req.trace else None
+        chosen: Optional[str] = None
+        if self._config.policy_selector and tenant_id:
+            try:
+                chosen = self._config.policy_selector(
+                    tenant_id=req.trace.tenant_id,
+                    capability=capability,
+                    preferred=req.preferred_model,
+                )
+            except Exception:  # noqa: BLE001
+                log.exception("ai_policy_selector_failed")
+
+        chosen = chosen or self._config.default_provider
+        return self._registry.get(chosen)
+
+    def _emit_usage(self, record: UsageRecord) -> None:
+        for cb in self._usage_callbacks:
+            try:
+                cb(record)
+            except Exception:  # noqa: BLE001
+                log.exception("ai_usage_callback_failed", provider=record.provider)
+
+    # ------------------------------------------------------------------
+    # Capability methods
+    # ------------------------------------------------------------------
+
+    async def chat(self, req: ChatRequest) -> ChatResponse:
+        provider = self._route(req, "chat")
+        log.debug("ai_chat_start", provider=provider.name, tenant=str(req.trace.tenant_id))
+        resp = await provider.chat(req)
+        self._emit_usage(
+            UsageRecord(
+                tenant_id=req.trace.tenant_id,
+                project_id=req.trace.project_id,
+                job_id=req.trace.job_id,
+                execution_id=resp.execution_id,
+                provider=resp.provider,
+                model=resp.model_used,
+                capability="chat",
+                input_tokens=resp.input_tokens,
+                output_tokens=resp.output_tokens,
+                estimated_cost=resp.estimated_cost,
+                duration_ms=resp.duration_ms,
+                recorded_at=datetime.now(tz=timezone.utc),
+            )
+        )
+        return resp
+
+    async def analyze_code(self, req: CodeAnalysisRequest) -> CodeAnalysisResult:
+        provider = self._route(req, "analyze_code")
+        resp = await provider.analyze_code(req)
+        self._emit_usage(
+            UsageRecord(
+                tenant_id=req.trace.tenant_id,
+                project_id=req.trace.project_id,
+                job_id=req.trace.job_id,
+                execution_id=resp.execution_id,
+                provider=resp.provider,
+                model=resp.model_used,
+                capability="analyze_code",
+                input_tokens=resp.input_tokens,
+                output_tokens=resp.output_tokens,
+                estimated_cost=resp.estimated_cost,
+                duration_ms=resp.duration_ms,
+                recorded_at=datetime.now(tz=timezone.utc),
+            )
+        )
+        return resp
+
+    async def generate_tests(self, req: TestGenerationRequest) -> GeneratedTests:
+        provider = self._route(req, "generate_tests")
+        resp = await provider.generate_tests(req)
+        self._emit_usage(
+            UsageRecord(
+                tenant_id=req.trace.tenant_id,
+                project_id=req.trace.project_id,
+                job_id=req.trace.job_id,
+                execution_id=resp.execution_id,
+                provider=resp.provider,
+                model=resp.model_used,
+                capability="generate_tests",
+                input_tokens=resp.input_tokens,
+                output_tokens=resp.output_tokens,
+                estimated_cost=resp.estimated_cost,
+                duration_ms=resp.duration_ms,
+                recorded_at=datetime.now(tz=timezone.utc),
+            )
+        )
+        return resp
+
+    async def decompose_requirement(
+        self, req: RequirementDecomposeRequest
+    ) -> RequirementDecomposition:
+        provider = self._route(req, "decompose_requirement")
+        resp = await provider.decompose_requirement(req)
+        self._emit_usage(
+            UsageRecord(
+                tenant_id=req.trace.tenant_id,
+                project_id=req.trace.project_id,
+                job_id=req.trace.job_id,
+                execution_id=resp.execution_id,
+                provider=resp.provider,
+                model=resp.model_used,
+                capability="decompose_requirement",
+                input_tokens=resp.input_tokens,
+                output_tokens=resp.output_tokens,
+                estimated_cost=resp.estimated_cost,
+                duration_ms=resp.duration_ms,
+                recorded_at=datetime.now(tz=timezone.utc),
+            )
+        )
+        return resp
+
+
+def build_default_gateway() -> AIGateway:
+    """Default bootstrap-configured gateway — uses MockProvider, stdout usage logging."""
+    gw = AIGateway()
+    gw.on_usage(lambda record: log.debug("ai_usage_record", **record.model_dump(mode="json")))
+    return gw

services/asf-ai/app/providers.py

@@ -0,0 +1,256 @@
+"""AIProvider interface + mock + registry."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Protocol, Optional, Type
+from uuid import uuid4, UUID
+
+import structlog
+
+from .capabilities import (
+    CapabilityRequest,
+    CapabilityResponse,
+    ChatRequest,
+    ChatResponse,
+    CodeAnalysisRequest,
+    CodeAnalysisResult,
+    TestGenerationRequest,
+    GeneratedTests,
+    RequirementDecomposeRequest,
+    RequirementDecomposition,
+    SubTask,
+    Finding,
+)
+
+log = structlog.get_logger(__name__)
+
+
+@dataclass
+class ProviderCaps:
+    supports_chat: bool = True
+    supports_code_analysis: bool = True
+    supports_test_generation: bool = True
+    supports_decompose: bool = True
+    max_input_chars: int = 200_000
+
+
+class AIProvider(Protocol):
+    """Low-level provider interface. The Gateway calls exactly one per request.
+
+    Concrete implementations: MockProvider, OpenAICompatProvider, NVIDIADGXProvider, ...
+    """
+
+    @property
+    def name(self) -> str: ...
+    @property
+    def default_model(self) -> str: ...
+    @property
+    def capabilities(self) -> ProviderCaps: ...
+
+    async def chat(self, req: ChatRequest) -> ChatResponse: ...
+    async def analyze_code(self, req: CodeAnalysisRequest) -> CodeAnalysisResult: ...
+    async def generate_tests(self, req: TestGenerationRequest) -> GeneratedTests: ...
+    async def decompose_requirement(
+        self, req: RequirementDecomposeRequest
+    ) -> RequirementDecomposition: ...
+
+
+# ==============================================================================
+# MockProvider — deterministic, canned, latency-simulating responses
+# ==============================================================================
+
+
+class MockProvider:
+    """Provider that always returns deterministic, predictable responses.
+
+    Uses:
+      * unit tests of code that depends on the AI Gateway
+      * local dev when no API keys are set
+      * UI demo / smoke tests
+    """
+
+    name = "mock"
+    default_model = "asf-mock-v1"
+    capabilities = ProviderCaps()
+
+    def __init__(self, *, response_latency_ms: int = 40) -> None:
+        self._latency = max(0, response_latency_ms)
+
+    async def _maybe_sleep(self) -> None:
+        if self._latency <= 0:
+            return
+        import asyncio
+        await asyncio.sleep(self._latency / 1000.0)
+
+    async def chat(self, req: ChatRequest) -> ChatResponse:
+        await self._maybe_sleep()
+        # Mirror the last user message back as a demo reply — never misleadingly smart
+        last_user = next(
+            (m.content for m in reversed(req.messages) if m.role == "user"),
+            "(no user message)",
+        )
+        reply = (
+            "[ASF Mock AI] This is a deterministic response from MockProvider. "
+            f"Your last message was: {last_user[:160]!r}. "
+            "To enable real AI, configure a non-mock provider in the ASF AI Gateway."
+        )
+        return ChatResponse(
+            provider=self.name,
+            model_used=self.default_model,
+            execution_id=uuid4(),
+            reply=reply,
+            references=["mock://no-real-data"],
+            input_tokens=len(last_user) // 4,
+            output_tokens=len(reply) // 4,
+            duration_ms=self._latency,
+            estimated_cost=0.0,
+        )
+
+    async def analyze_code(self, req: CodeAnalysisRequest) -> CodeAnalysisResult:
+        await self._maybe_sleep()
+        lines = [ln for ln in req.code_snippet.splitlines() if ln.strip()]
+        findings: list[Finding] = []
+        if "TODO" in req.code_snippet:
+            findings.append(
+                Finding(
+                    severity="info",
+                    kind="todo",
+                    message="Mock analysis found TODO comment — remember to follow up.",
+                )
+            )
+        if "except Exception" in req.code_snippet:
+            findings.append(
+                Finding(
+                    severity="medium",
+                    kind="broad_exception",
+                    message="Broad except clause detected — consider narrowing exception types.",
+                )
+            )
+        return CodeAnalysisResult(
+            provider=self.name,
+            model_used=self.default_model,
+            execution_id=uuid4(),
+            summary=f"Mock: analyzed {len(lines)} non-empty lines; {len(findings)} informational finding(s).",
+            findings=findings,
+            input_tokens=max(10, len(req.code_snippet) // 4),
+            output_tokens=100,
+            duration_ms=self._latency,
+            estimated_cost=0.0,
+        )
+
+    async def generate_tests(self, req: TestGenerationRequest) -> GeneratedTests:
+        await self._maybe_sleep()
+        test_cases = [
+            "# Generated by MockProvider — placeholder skeleton\n"
+            "def test_mock_happy_path():\n"
+            "    # TODO: replace with real assertions from a production AI provider\n"
+            "    assert True\n",
+            "# Generated by MockProvider — placeholder skeleton\n"
+            "def test_mock_failure_mode():\n"
+            "    # TODO: simulate and assert on failure behavior\n"
+            "    assert True\n",
+        ]
+        return GeneratedTests(
+            provider=self.name,
+            model_used=self.default_model,
+            execution_id=uuid4(),
+            test_cases=test_cases,
+            fixtures_needed=[
+                "NOTE: Mock provider — fixtures listed are placeholders only.",
+            ],
+            coverage_notes="Mock response. Real AI provider will target specific branches.",
+            input_tokens=max(10, len(req.code_snippet) // 4),
+            output_tokens=300,
+            duration_ms=self._latency,
+            estimated_cost=0.0,
+        )
+
+    async def decompose_requirement(
+        self, req: RequirementDecomposeRequest
+    ) -> RequirementDecomposition:
+        await self._maybe_sleep()
+        limit = max(1, min(5, req.max_subtasks))
+        subtasks = [
+            SubTask(
+                title=f"(Mock) Clarify scope of requirement",
+                description="Mock placeholder — confirm acceptance criteria with stakeholders.",
+                estimated_complexity="s",
+            ),
+            SubTask(
+                title=f"(Mock) Draft architecture",
+                description="Mock placeholder — identify modules, data flow, contracts.",
+                estimated_complexity="m",
+                depends_on=[0],
+            ),
+            SubTask(
+                title=f"(Mock) Implement feature",
+                description="Mock placeholder — concrete coding work.",
+                estimated_complexity="l",
+                depends_on=[1],
+            ),
+            SubTask(
+                title=f"(Mock) Add tests",
+                description="Mock placeholder — unit + integration coverage.",
+                estimated_complexity="m",
+                depends_on=[2],
+            ),
+            SubTask(
+                title=f"(Mock) Deploy + monitor",
+                description="Mock placeholder — staging → production, observability.",
+                estimated_complexity="s",
+                depends_on=[3],
+            ),
+        ][:limit]
+        return RequirementDecomposition(
+            provider=self.name,
+            model_used=self.default_model,
+            execution_id=uuid4(),
+            subtasks=subtasks,
+            architecture_notes=["(Mock) Decomposition is deterministic placeholder data."],
+            open_questions=[
+                "(Mock) Which non-functional requirements apply?",
+                "(Mock) Any existing integration contracts to preserve?",
+            ],
+            input_tokens=max(10, len(req.requirement_text) // 4),
+            output_tokens=400,
+            duration_ms=self._latency,
+            estimated_cost=0.0,
+        )
+
+
+# ==============================================================================
+# ProviderRegistry — maps provider name → AIProvider instance
+# ==============================================================================
+
+
+class ProviderRegistry:
+    def __init__(self) -> None:
+        self._providers: dict[str, AIProvider] = {}
+
+    def register(self, name: str, provider: AIProvider) -> None:
+        self._providers[name] = provider
+
+    def get(self, name: str) -> AIProvider:
+        if name not in self._providers:
+            raise KeyError(f"Unknown AI provider: {name}. Registered: {list(self._providers)}")
+        return self._providers[name]
+
+    def names(self) -> list[str]:
+        return list(self._providers.keys())
+
+
+def default_registry() -> ProviderRegistry:
+    r = ProviderRegistry()
+    r.register("mock", MockProvider(response_latency_ms=40))
+    return r
+
+
+def build_default_gateway():
+    """Instantiate the canonical production AIGateway (mock by default).
+
+    Imported lazily to avoid a cycle with ``app.gateway``.
+    """
+    from .gateway import AIGateway, AIGatewayConfig
+
+    return AIGateway(config=AIGatewayConfig(), registry=default_registry())

services/asf-ai/app/usage.py

@@ -0,0 +1,25 @@
+"""Usage record (for billing / metering later)."""
+
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Optional
+from uuid import UUID
+
+from pydantic import BaseModel
+
+
+class UsageRecord(BaseModel):
+    tenant_id: UUID
+    project_id: Optional[UUID] = None
+    job_id: Optional[UUID] = None
+    execution_id: UUID
+    provider: str
+    model: str
+    capability: str
+    input_tokens: Optional[int] = None
+    output_tokens: Optional[int] = None
+    gpu_seconds: Optional[float] = None
+    estimated_cost: Optional[float] = None
+    duration_ms: Optional[int] = None
+    recorded_at: datetime