feat: add durable v2 memory learning and tasks

Author: EvanProgramming

Makefile

@@ -2,6 +2,7 @@
 
 # Use a known-stable Python version (3.14 has import deadlocks with openai)
 PYTHON := python3.12
+VENV_PYTHON := $(if $(wildcard venv/bin/python),./venv/bin/python,$(PYTHON))
 
 # Detect Windows (native cmd) and redirect to .bat files
 ifeq ($(OS),Windows_NT)
@@ -18,11 +19,11 @@ install:
 
 run:
 	@command -v $(PYTHON) >/dev/null 2>&1 || { echo "Error: venv requires $(PYTHON). Run 'make install' with $(PYTHON) installed."; exit 1; }
-	. venv/bin/activate && python main.py
+	$(VENV_PYTHON) main.py
 
 web:
 	@command -v $(PYTHON) >/dev/null 2>&1 || { echo "Error: venv requires $(PYTHON)."; exit 1; }
-	. venv/bin/activate && pip install fastapi uvicorn -q && python server.py
+	$(VENV_PYTHON) -m pip install fastapi uvicorn -q && $(VENV_PYTHON) server.py
 
 debug:
 	@command -v $(PYTHON) >/dev/null 2>&1 || { echo "Error: venv requires $(PYTHON). Run 'make install' first."; exit 1; }
@@ -44,23 +45,21 @@ clean:
 
 # Syntax check
 lint:
-	. venv/bin/activate && python -c "compile(open('main.py').read(), 'main.py', 'exec'); print('main.py OK')"
-	. venv/bin/activate && python -c "compile(open('tools.py').read(), 'tools.py', 'exec'); print('tools.py OK')"
-	. venv/bin/activate && python -c "compile(open('memory.py').read(), 'memory.py', 'exec'); print('memory.py OK')"
+	$(VENV_PYTHON) -m compileall -q main.py server.py tools.py memory.py event_store.py task_engine.py learning_engine.py migration.py
+	@echo "Python syntax OK."
 	@echo "All files pass syntax check."
 
 # Unit tests
 test:
-	. venv/bin/activate && pip install fastapi uvicorn -q && python -m unittest discover -s tests -p 'test_*.py' -v
+	$(VENV_PYTHON) -m unittest discover -s tests -p 'test_*.py' -v
 
 # Quick verification
 check:
 	@echo "Checking Python syntax..."
-	@./venv/bin/python -c "compile(open('main.py').read(), 'main.py', 'exec'); print('  main.py: OK')"
-	@./venv/bin/python -c "compile(open('tools.py').read(), 'tools.py', 'exec'); print('  tools.py: OK')"
-	@./venv/bin/python -c "compile(open('memory.py').read(), 'memory.py', 'exec'); print('  memory.py: OK')"
+	@$(VENV_PYTHON) -m py_compile main.py server.py tools.py memory.py event_store.py task_engine.py learning_engine.py migration.py
+	@echo "  Python modules: OK"
 	@echo "Checking git tools..."
-	@./venv/bin/python -c "from tools import AVAILABLE_TOOLS; git = [k for k in AVAILABLE_TOOLS if k.startswith('git_')]; print(f'  {len(git)} git tools, {len(AVAILABLE_TOOLS)} total tools')"
+	@$(VENV_PYTHON) -c "from tools import AVAILABLE_TOOLS; git = [k for k in AVAILABLE_TOOLS if k.startswith('git_')]; print(f'  {len(git)} git tools, {len(AVAILABLE_TOOLS)} total tools')"
 	@echo "All checks passed."
 
 # Git helpers

README.md

@@ -60,7 +60,7 @@
 OpenKyrozen is a **self-learning AI agent** that runs in your terminal. Unlike a typical chatbot, it:
 
 - **Uses 26 built-in tools** — read/write files, execute shell commands, search the web, manage git repositories
-- **Learns continuously** — 20 self-learning features run in the background, extracting facts, inventing skills, and improving strategies
+- **Learns continuously** — background learning creates evidence-backed proposals and only promotes repeated or validated improvements
 - **Works with any LLM** — DeepSeek, OpenAI, Claude, Gemini, or local Ollama models
 - **Runs on any OS** — macOS, Linux, and Windows (with automatic terminal capability detection)
 - **Has a Web UI** — browser-based chat interface with REST API for integration
@@ -341,11 +341,11 @@ For multi-step work (refactors, project generators, codebase audits):
 
 ## 🧬 Self-Learning System
 
-This is what makes Kyrozen different. **20 self-learning features** run continuously in the background — no manual saving needed. The agent gets smarter the longer you use it.
+The v2 learning engine is autonomous but evidence-gated. It records observations, creates candidate proposals, checks repeated evidence or validation results, activates a versioned improvement, and records rollback information. Autonomous learning never grants a new permission merely because a memory contains an instruction.
 
 ### How it works
 
-Every 30 seconds (when you're idle), Kyrozen runs a learning cycle. Most features can be toggled on/off with `/self-learning`; the remaining nine run automatically in the background.
+When idle, Kyrozen runs a bounded learning cycle. File scans are incremental, background work has a concurrency limit, and failures are recorded as events instead of being silently discarded. Most features can be toggled on/off with `/self-learning`.
 
 | # | Feature | What it learns |
 |---|---------|---------------|
@@ -372,7 +372,21 @@ Every 30 seconds (when you're idle), Kyrozen runs a learning cycle. Most feature
 
 ### Memory storage
 
-Long-term memory uses **ChromaDB** (vector database, stored in `chroma_memory/`). Falls back to in-memory storage if ChromaDB is unavailable. Memories are semantically searchable — the agent can recall relevant facts from weeks ago.
+OpenKyrozen v2 uses **SQLite as the source of truth** (`~/.kyrozen/v2/openkyrozen.sqlite3`) and ChromaDB as a rebuildable semantic index. Memories have a kind, scope, confidence, source events, and lifecycle status. Workspaces and sessions are isolated, raw observations are marked as data, and `/forget` removes records by durable ID. If ChromaDB is unavailable, SQLite keeps durable keyword retrieval.
+
+Import an existing v1 store without deleting it:
+
+```bash
+python main.py migrate v1 ./chroma_memory
+```
+
+The migration creates a `.v1-backup` copy and writes the v2 database under `~/.kyrozen/v2/` (or `KYROZEN_DB_PATH`).
+
+### v2 durable tasks and learning
+
+Tasks persist across process restarts and use `pending`, `running`, `succeeded`, `failed`, `blocked`, and `cancelled` states. `TaskDone` is only a completion request; a successful tool result, test, file check, or explicit confirmation must provide evidence before a task can succeed.
+
+Learning proposals are visible with `/learning status`, explainable with `/learning explain <proposal_id>`, and reversible with `/learning rollback <proposal_id>`.
 
 ---
 
@@ -395,6 +409,11 @@ KYROZEN_SERVER_TOKEN=change-me python server.py --host 0.0.0.0 --port 8000
 | `POST` | `/api/chat` | Send a message, get JSON response |
 | `POST` | `/api/chat/stream` | SSE streaming chat |
 | `GET` | `/api/memory?q=keyword` | Search stored memories |
+| `GET` | `/api/v2/memory?q=keyword` | Structured memory search with provenance and scope |
+| `GET/POST` | `/api/v2/tasks` | Durable task listing and creation |
+| `GET` | `/api/v2/learning` | Learning proposal status |
+| `POST` | `/api/v2/learning/{id}/rollback` | Roll back an activated proposal |
+| `GET` | `/api/v2/events` | Auditable runtime, task, session, and learning events |
 | `GET` | `/api/cost` | Token usage and cost summary |
 | `GET` | `/api/health` | Provider status + memory count |
 | `GET` | `/api/voice/speak?text=...` | Text-to-speech via system TTS |

README.zh-CN.md

@@ -341,11 +341,11 @@ python main.py
 
 ## 🧬 自学习系统
 
-这是 Kyrozen 与众不同的地方。**20 项自学习功能**在后台持续运行——无需手动保存。使用时间越长，智能体就越聪明。
+这是 Kyrozen 与众不同的地方。v2 的自学习是自动的，但必须经过证据门控：系统先记录观察，再生成候选提案，经过重复证据或验证后才激活，并保留回滚信息。记忆中的文字永远不能自动授予新权限。
 
 ### 工作原理
 
-每 30 秒（当你空闲时），Kyrozen 运行一个学习周期。大部分功能可以通过 `/self-learning` 开关；其余九项在后台自动运行。
+空闲时 Kyrozen 运行有界的学习周期。项目扫描采用增量方式，后台任务有并发限制，失败会记录为事件而不会静默丢弃。大部分功能可以通过 `/self-learning` 开关。
 
 | # | 功能 | 学习内容 |
 |---|------|---------|
@@ -372,7 +372,17 @@ python main.py
 
 ### 记忆存储
 
-长期记忆使用 **ChromaDB**（向量数据库，存储在 `chroma_memory/`）。如果 ChromaDB 不可用则回退到内存存储。记忆支持语义搜索——智能体可以回忆起几周前的相关事实。
+OpenKyrozen v2 使用 **SQLite 作为事实主库**（`~/.kyrozen/v2/openkyrozen.sqlite3`），ChromaDB 只作为可重建的语义索引。记忆包含类型、作用域、置信度、来源事件和生命周期状态；workspace 与 session 相互隔离。即使 ChromaDB 不可用，SQLite 仍会提供持久化关键词检索。
+
+导入旧版 Chroma 记忆而不删除原数据：
+
+```bash
+python main.py migrate v1 ./chroma_memory
+```
+
+该命令会创建 `.v1-backup` 备份，并将 v2 数据写入 `~/.kyrozen/v2/`（也可通过 `KYROZEN_DB_PATH` 指定）。
+
+任务会跨重启保存。`TaskDone` 只是完成请求，只有工具结果、测试、文件检查或明确确认提供证据后，任务才会进入成功状态。可使用 `/learning status`、`/learning explain <proposal_id>` 和 `/learning rollback <proposal_id>` 管理学习提案。
 
 ---
 
@@ -395,6 +405,11 @@ KYROZEN_SERVER_TOKEN=change-me python server.py --host 0.0.0.0 --port 8000
 | `POST` | `/api/chat` | 发送消息，获取 JSON 响应 |
 | `POST` | `/api/chat/stream` | SSE 流式聊天 |
 | `GET` | `/api/memory?q=关键词` | 搜索已存储的记忆 |
+| `GET` | `/api/v2/memory?q=关键词` | 返回带来源、置信度和作用域的结构化记忆 |
+| `GET/POST` | `/api/v2/tasks` | 持久化任务查询与创建 |
+| `GET` | `/api/v2/learning` | 查看学习提案 |
+| `POST` | `/api/v2/learning/{id}/rollback` | 回滚已激活的学习提案 |
+| `GET` | `/api/v2/events` | 查看运行时、会话、任务和学习审计事件 |
 | `GET` | `/api/cost` | Token 用量和费用摘要 |
 | `GET` | `/api/health` | 服务商状态 + 记忆计数 |
 | `GET` | `/api/voice/speak?text=...` | 通过系统 TTS 进行文本转语音 |
@@ -406,8 +421,8 @@ KYROZEN_SERVER_TOKEN=change-me python server.py --host 0.0.0.0 --port 8000
 
 API 和 MCP 路由在本机回环访问时可以不使用令牌；任何非本机部署都必须
 设置 `KYROZEN_SERVER_TOKEN`，并通过 `Authorization: Bearer <token>` 或
-`X-Kyrozen-Token` 发送。MCP 的写文件、Shell、Git 修改和远程克隆工具默认
-关闭；只有在可信环境中才应显式设置 `KYROZEN_MCP_ALLOW_DANGEROUS=1`。
+`X-Kyrozen-Token` 发送。MCP/Web 默认使用 `workspace` 能力，`full` 才会开放
+不可逆 Git reset 和动态 Python 工具；高影响 Git 操作仍受确认模式保护。
 
 ### Docker 部署