ltcai 8.9.0 → 9.0.0

package/docs/ROADMAP_RECOMMENDATIONS.md

@@ -5,43 +5,12 @@ small release-sized work. The operating principle stays unchanged:
 
 > Models are temporary. Knowledge is durable. The Brain is the product.
 
-## 7.3.0 Applied Slice
+## Applied Release History
 
-7.3.0 implements the first narrow slice of the roadmap:
-
-- Runtime evolution: single-agent and multi-agent execution share
-  `agent-run-contract/v1`, making mode/status/timeline evidence inspectable.
-- Hybrid search optimization: the Brain quality gate now includes deterministic
-  recall and ranking regression thresholds.
-- Security and trust: run contracts distinguish runtime type and execution mode
-  so simulated output is not presented as real execution.
-
-## 7.4.0 Applied Slice
-
-7.4.0 completes the next roadmap slice without deferring it:
-
-- Runtime convergence: agent runs, workflow runs, audit events, and realtime
-  events all expose the `agent-run-contract/v1` family envelope while keeping
-  legacy top-level fields for compatibility.
-- Trust and operations: persisted run rows refresh their contract through
-  queued, running, terminal, cancelled, and interrupted states; audit events are
-  contracted only after secret redaction.
-- Retrieval quality and scale: the CI quality gate now seeds a real local
-  Knowledge Graph corpus and scores SearchService hybrid retrieval with judged
-  queries, recall, precision, NDCG, and must-include hit-rate thresholds.
-
-## 7.5.0 Applied Slice
-
-7.5.0 burns down the remaining 7.4.0 risk instead of deferring it:
-
-- Contract consumption: AgentRuntime and realtime feed APIs now emit compact
-  `contracts` views so UI, replay, admin, and export consumers can depend on the
-  shared family envelope directly.
-- Retrieval scale: the corpus fixture now runs against 250+ local records while
-  keeping judged queries, graded relevance, and must-include expectations.
-- Release trust: stale artifact mixing is handled through clean exact-version
-  artifact generation, npm audit findings are cleared, and the Tauri 2 stack is
-  updated past the old `block v0.1.6` future-incompatibility warning.
+Release-specific applied-slice history is now kept only for 8.0.0 and newer in
+[docs/CHANGELOG.md](CHANGELOG.md) and [RELEASE.md](../RELEASE.md). Pre-8.0.0
+release-history notes were removed from the tracked documentation surface during
+the 9.0.0 documentation cleanup.
 
 ## Near-Term Tracks
 
@@ -79,3 +48,59 @@ small release-sized work. The operating principle stays unchanged:
 - Interoperability with Obsidian, Notion, Email, Calendar, Git, Slack, and Teams.
 - Encrypted Brain Network sharing.
 - Plugin marketplace and public benchmarks.
+
+## Large-Scale Work Feasibility Assessment (2026-07-07)
+
+2026-07-07 코드베이스 검사 + 테스트 실행 결과 기반. (app_factory 1214줄, workspace_os 1391줄, 최근 AgentRuntime facade + KG mixin 분해 완료, 833 unit tests passed)
+
+### 1. Knowledge Graph / Retrieval at Large Corpus Scale (가장 추천, 기반 탄탄)
+- Incremental/background ingestion + advanced dedup(콘텐츠해시 이상), conflict resolution, merge.
+- Larger corpus benchmarks (현재 250+ → 수천~수만 아이템), latency budget, channel별 진단.
+- pgvector 풀 프로덕션 경로 + 마이그레이션.
+- **실행 가능성**: IngestionPipeline + provenance + vector mixin + benchmark fixtures 이미 존재. discovery/retrieval 분해 완료. hooks lifecycle 통합됨. 대형 작업으로 적합. 테스트: test_ingestion_pipeline.py 14/14 green.
+
+### 2. Multi-Modal Brain (스키마 준비됨, 구현 공백 큼)
+- 이미지: vision LLM describe + 임베딩, IMAGE/IMAGE_TEXT 노드 + CONTAINS_IMAGE 엣지 완성, UI 증거 표시.
+- 오디오(전사), 비디오(키프레임).
+- **실행 가능성**: schema.py에 IMAGE/IMAGE_TEXT/CONTAINS_IMAGE 정의됨. discovery_index.py에 PIL + pytesseract OCR (include_ocr) 부분 구현. embeddings.py / LLMRouter에 vision 지원 부재. retrieval도 이미지 노드 미처리. ingestion 라우팅 확장 필요. 대형 + 차별화 포인트.
+
+### 3. Server / Runtime Composition 완전화 (구조 부채 해소)
+- app_factory.py 잔여 로직 추가 추출 (더 많은 wiring → runtime/* 전용 모듈).
+- dict(locals()) 레거시 제거, 강력한 RuntimeContext / DI.
+- Config 중앙화 감사 및 강제 (core/config.py 이미 중앙).
+- **실행 가능성**: 최근 review0707 + decomp (1214줄로 감소, runtime/ 디렉토리 풍부) 후에도 여전히 1.2k 라인. AGENTS.md 우선순위 #3,4 정확히 매치. AgentRuntime은 lattice_brain/runtime/agent_runtime.py 로 facade 추출 완료. ToolRegistry도 core/ + services 분리 진행.
+
+### 4. Proactive Synthesis + Temporal / Contradiction (최근 작업 연장)
+- 백그라운드 합성 잡, 그래프 이력 기반 모순 검출, temporal 쿼리.
+- 최근 커밋(Brain synthesis memories, follow-up) 을 스케줄 + 자동화로 확장.
+- **실행 가능성**: MultiAgentOrchestrator + workflow_engine + IngestionPipeline + Brain Brief 이미 wired. hooks/audit 있음. agent synthesis UI 표면 최근 추가. 대형 기능으로 자연스러운 확장.
+
+### 5. Ecosystem / Interop + Plugin Marketplace (외부 연동 대형)
+- Obsidian/Notion/Email/Calendar import bridge (MCP 또는 전용 ingestion).
+- 서명된 플러그인, 원격 카탈로그, marketplace UI.
+- Brain Network (암호화 공유).
+- **실행 가능성**: plugins/ (hello-world, git-insights), mcp_registry, marketplace.py, core/plugins.py 존재. ingestion이 단일 관문. 하지만 실제 외부 브릿지/마켓플레이스 UI/보안 스캔은 대형.
+
+기타: Tauri auto-update + native scale, Workflow Designer 프론트엔드 완성, VSCode 확장 심화 통합 등.
+
+**검증 결과 (2026-07-07)**:
+- Core imports (Config, IngestionPipeline, AgentRuntime, ToolRegistry): OK
+- Unit tests: 833 passed (tests/unit/)
+- Ingestion specific: 14 passed
+- Agent runtime tests: 29 passed
+- Frontend lint: typecheck + privacy + OpenAPI 335 paths + lint all pass
+- Git: clean (inspection 시점)
+
+**2026-07-07 pts_grok 대형 후보 1~5 슬라이스 실행 결과 (자율 진행)**:
+- 1. KG/Retrieval: background ingestion queue seam, schedule_background, vector index scale diagnostics, backlog reasons/samples, coverage ratio, and rebuild latency budget.
+- 2. Multimodal: offline VisionStub describe/embed path and discovery_index vision_caption fallback so image files have searchable evidence without remote calls.
+- 3. Server decomp: explicit _RUNTIME_BUNDLE in app_factory as the migration contract toward DI while dict(locals()) compatibility remains.
+- 3a. Server decomp follow-up: filtered app_factory runtime namespace so internal scratch imports/runtime dicts no longer leak through lazy server_app compatibility; namespace reduced from 300 to 253 keys while legacy helpers remain, with typed RuntimeBundle added behind legacy _RUNTIME_BUNDLE.
+- 4. Proactive/Temporal: stronger MemoryQualityManager conflict detection, including pairwise opposite-preference and temporal-negation flags.
+- 4a. Proactive UX follow-up: Brain Brief now emits proactive_actions, and BrainHome renders one-click ask/delegate/review/graph action cards.
+- 4b. Proactive UX completion: BrainHome now keeps a local activity trail for suggested actions, showing whether Brain ask, Agent delegation, Review draft, or graph navigation actions are running, completed, or failed.
+- 5. Interop/Marketplace: ingestion_bridge marketplace templates and /marketplace/interop/bridges exposure for future Obsidian/Calendar-style connector imports through unified-ingestion.
+- 총 affected tests: 45 passed in the first targeted gate. Broader lint/static/doc checks are run before commit.
+- 남은: 각 슬라이스는 기반 확장. full background workers, real local VLM/image embeddings, full locals() removal, graph-level temporal queries, and production connector installs remain follow-up work.
+
+이 항목들은 "대형사이즈" 에도 할 만하며, mission (local-first, KG, AgentRuntime, security) 과 AGENTS 우선 리팩토링 순서에 부합. 작은 슬라이스로 시작 추천.

package/docs/TRUST_MODEL.md

@@ -1,6 +1,6 @@
 # Lattice AI Trust Model
 
-Current release: **8.9.0 — Scoped Memory & Tool Policy Hardening**.
+Current release: **9.0.0 — Code Review Closure & Runtime Cleanup**.
 
 Lattice AI is local-first, explicit about external communication, and honest
 when a capability is unavailable.

package/docs/WHY_LATTICE.md

@@ -1,6 +1,6 @@
 # Why Lattice AI Exists
 
-Current release: **8.9.0 — Scoped Memory & Tool Policy Hardening**.
+Current release: **9.0.0 — Code Review Closure & Runtime Cleanup**.
 
 **Lattice AI is a local-first Digital Brain that keeps your knowledge durable
 across any AI model.**
@@ -32,7 +32,7 @@ The graph matters, but it is not the product identity. The product identity is a
 local-first Digital Brain: a place where the user can talk, add sources, watch
 memory grow, and inspect the Knowledge Graph when they need proof.
 
-In 8.9.0 the first screen is intentionally not a dashboard. The living Brain,
+In 9.0.0 the first screen is intentionally not a dashboard. The living Brain,
 conversation composer, evidence-backed Brain Brief, scoped memory isolation,
 and five-minute onboarding
 path appear together so the user immediately knows what matters, what to add,

package/docs/kg-schema.md

@@ -1,6 +1,6 @@
 # Knowledge Graph Schema
 
-Current release: **8.9.0 — Scoped Memory & Tool Policy Hardening**.
+Current release: **9.0.0 — Code Review Closure & Runtime Cleanup**.
 
 명세 출처: `lattice_ai_full_spec.pptx` 슬라이드 20·21·22
 구현: `kg_schema.py`

package/lattice_brain/__init__.py

@@ -26,7 +26,7 @@ from .storage import (
     storage_from_env,
 )
 
-__version__ = "8.9.0"
+__version__ = "9.0.0"
 
 __all__ = [
     "AgentRuntime",

package/lattice_brain/archive.py

@@ -9,7 +9,6 @@ restore on another machine without contacting a service.
 from __future__ import annotations
 
 import base64
-import hashlib
 import io
 import json
 import os
@@ -27,6 +26,8 @@ from cryptography.hazmat.primitives.ciphers.aead import AESGCM
 from cryptography.exceptions import InvalidTag
 from cryptography.hazmat.primitives.kdf.pbkdf2 import PBKDF2HMAC
 
+from .utils import sha256_file as _sha256_file
+
 
 ARCHIVE_FORMAT = "latticebrain.encrypted"
 ARCHIVE_VERSION = 2
@@ -68,15 +69,9 @@ def _derive_key(passphrase: str, salt: bytes) -> bytes:
 
 
 def _sha256_bytes(data: bytes) -> str:
-    return hashlib.sha256(data).hexdigest()
+    import hashlib
 
-
-def _sha256_file(path: Path) -> str:
-    h = hashlib.sha256()
-    with open(path, "rb") as fh:
-        for block in iter(lambda: fh.read(65536), b""):
-            h.update(block)
-    return h.hexdigest()
+    return hashlib.sha256(data).hexdigest()
 
 
 def _safe_json(value: Any) -> Any:

package/lattice_brain/embeddings.py

@@ -8,7 +8,7 @@ import os
 import re
 import struct
 from dataclasses import dataclass
-from typing import Iterable, List
+from typing import Any, Dict, Iterable, List, Optional
 
 
 DEFAULT_EMBEDDING_DIM = int(os.getenv("LATTICEAI_VECTOR_DIM", "384"))
@@ -87,4 +87,40 @@ class LocalEmbeddingModel:
         return list(struct.unpack(f"<{count}f", payload[: count * 4]))
 
 
-__all__ = ["DEFAULT_EMBEDDING_DIM", "EMBEDDING_MODEL_ID", "LocalEmbeddingModel", "embedding_model_id"]
+# --- Large candidate #2 slice: multimodal (vision) stubs ---
+# Schema already defines IMAGE / IMAGE_TEXT / CONTAINS_IMAGE.
+# These stubs allow ingestion + retrieval paths to carry image signals without
+# requiring heavy deps at core. Real impl can swap in local vision (e.g. via
+# ollama vision or onnx CLIP) behind the same interface.
+@dataclass(frozen=True)
+class VisionStub:
+    """Offline vision describe + embed stubs for multimodal Brain.
+
+    describe: returns a short textual caption derived from metadata/filename.
+    embed: produces a vector from image meta (size+format) + optional caption hash.
+    Later: replace with real embedding model that accepts image bytes/path.
+    """
+    dim: int = DEFAULT_EMBEDDING_DIM
+
+    def describe(self, path: str | None = None, meta: Optional[Dict[str, Any]] = None) -> str:
+        meta = meta or {}
+        w = meta.get("width") or meta.get("w") or "?"
+        h = meta.get("height") or meta.get("h") or "?"
+        fmt = meta.get("format") or meta.get("ext") or "img"
+        name = (path or "").split("/")[-1] or "image"
+        # deterministic caption stub (no external call)
+        return f"Image {name} ({fmt} {w}x{h})"
+
+    def embed_image(self, path: str | None = None, meta: Optional[Dict[str, Any]] = None, caption: str = "") -> List[float]:
+        meta = meta or {}
+        basis = f"{path or ''}|{meta.get('width',0)}x{meta.get('height',0)}|{meta.get('format','')}|{caption[:120]}"
+        # reuse text embedder for determinism (image content hash would be better with real vision)
+        model = LocalEmbeddingModel(dim=self.dim)
+        return model.embed(basis)
+
+
+def get_vision_embedder(dim: int | None = None) -> VisionStub:
+    return VisionStub(dim=dim or DEFAULT_EMBEDDING_DIM)
+
+
+__all__ = ["DEFAULT_EMBEDDING_DIM", "EMBEDDING_MODEL_ID", "LocalEmbeddingModel", "embedding_model_id", "VisionStub", "get_vision_embedder"]