Open-Sable

Open-Sable is an MIT-licensed autonomous AI agent framework from Ideoa Labs, written in Python. It runs continuously on a local machine, connects to chat platforms such as Telegram, Discord and Slack, and executes tasks through a library of tools and skills. The latest release is v1.7.0, published April 2026.

What It Is

Open-Sable is an agent framework that combines an LLM backend with cognitive subsystems: a goal system, layered memory, meta-learning, tool synthesis, a world model and metacognition. It is local-first, trying Ollama before falling back to any of 12 configurable cloud LLM providers, so data can stay on the user's machine. Each agent is a profile folder with its own soul.md identity, profile.env configuration and tools.json access rules, so several independent agents can run from one codebase.

How It Works

A recurring cognitive tick pipeline (71 phases in v1.7.0) drives memory consolidation, reflection, goal handling, social scanning, self-healing and more. Tool calls use a ReAct-style loop and a deep planner that breaks goals into step DAGs with re-planning on failure. Lazy tool loading compacts tool schemas for smaller models to save context. Tool access can be restricted per agent with allowlists or denylists.

Capabilities

• Chat interfaces: Telegram, Discord, WhatsApp, Slack, Matrix, IRC, email, CLI, mobile REST API and voice call.
• Social integrations for X, Instagram, Facebook, LinkedIn, YouTube and TikTok.
• RAG pipeline with ChromaDB, a workflow engine, and Word/Excel/PDF/PowerPoint document creation.
• Vision-based desktop control, Whisper speech-to-text and Piper text-to-speech.
• SkillFactory, which generates, validates and publishes new skills from plain-language descriptions.
• Multi-exchange trading with paper mode and risk limits.

Setup and Deployment

The project requires Python 3.11+ and 8GB RAM, with Ollama optional. It installs via an automated installer script or manual pip install, and a Raspberry Pi installer is provided. Docker Compose and Kubernetes manifests are included, and desktop installers are offered for Windows, macOS and Linux. The README lists tool synthesis and multi-device sync as experimental.