你那个真正自主、本地、永远属于你的个人AI。

Open-Sable 是一款下一代自主 AI 代理框架，配备代理 AI 认知子系统。它全天候运行于您的本地机器，集成您喜欢的信使，执行真实任务，并不断自我改进，同时保护您的数据隐私。

✅ 现在有效的方法

核心平台

• 本地优先运行时，具备持久的后台进程管理（start.sh）、日志、健康检查和基于配置文件的代理。
• 多代理架构，每个配置文件拥有隔离的身份、配置、工具权限、套接字、数据和内存。
• 安全工具执行、工作流引擎、目标系统、可审计性和自主恢复（看门狗/重启/热重载）。

LLM与认知

• Ollama-首推本地推理，带云服务提供商的备份。
• 低显存模式，支持稳健的备用路由和启动自动拉取（默认为 qwen3.5:0.8b。
• 认知记忆堆栈（工作/情节/长期）、衰减+巩固、反思和主动推理循环。

Open-Sable 🚀

Your personal AI that actually does things autonomous, local, and yours forever.

Open-Sable is a next-generation autonomous AI agent framework with Agentic AI cognitive subsystems. It runs 24/7 on your local machine, integrates with your favorite messengers, executes real-world tasks, and continuously improves itself, all while keeping your data private.

✅ What works right now

Core platform

• Local-first runtime with persistent background process management (start.sh), logs, health checks, and profile-based agents.
• Multi-agent architecture with isolated identity, config, tool permissions, sockets, data, and memory per profile.
• Safe tool execution, workflow engine, goal system, auditability, and autonomous recovery (watchdog/restart/hot-reload).

LLM and cognition

• Ollama-first local inference with cloud-provider fallback.
• Low-VRAM mode with robust fallback routing and startup auto-pull support (defaulting to qwen3.5:0.8b).
• Cognitive memory stack (working/episodic/long-term), decay + consolidation, reflection, and proactive reasoning loops.
• ReAct-style multi-step tool chaining, deep DAG planning with re-planning, and benchmark-driven self-improvement.

Skills and integrations

• Telegram interface, SkillFactory pipeline, RAG/document workflows, clipboard/OCR, and 21+ community skills.
• Productivity stack: SMTP/IMAP email + Google Calendar.
• Developer stack: full GitHub integration (issues, PRs, branches, code search, releases).
• Trading stack: multi-exchange trading support (crypto, stocks, prediction markets) with risk controls.

Recently verified (Mar 2026)

• Stable Zunvra registration/login with persisted credentials and restart-safe reconnect behavior.
• Correct profile sync path (avatar/bio/banner), with gateway path normalization fixes.
• Autonomous Zunvra social loop is active (posting/replies/engagement), including output cleanup safeguards.
• Image posting pipeline works end-to-end via compatibility payload mapping (imageUrls + media_urls).
• Better profile isolation for Zunvra intel/social flags and profile-scoped intel data directories.