symbiotic.sh

Category: AI/ML

Symbiotic.sh is a personal AI operating system designed for security-conscious developers and power users. It functions as a single, persistent daemon that creates a comprehensive, private memory of a user's digital life. The system is built around a continuous feedback loop: Capture → Intake → Distillery → Archive → Recall → Action → Evolution. It ingests unstructured content like notes, articles, and bookmarks, and its 'Distillery' pipeline uses AI to extract and structure this information into typed, queryable facts. These facts are stored in a locally encrypted 'Archive', which serves as the single source of truth for the entire system.

What sets Symbiotic apart is its sophisticated agent architecture. Instead of relying on simple tool-calling, it coordinates a swarm of specialist agents—such as a security auditor, architecture analyst, and risk adversary—that collaborate on complex tasks. These agents communicate by writing their findings back into the central Archive as structured data, creating a clean, retrievable, and auditable state for orchestration. This allows for more complex, multi-step reasoning than traditional agent frameworks. A 'Recall Gateway' intelligently packs context from semantic, episodic, and procedural memory spaces to provide agents with the precise information they need.

Security and privacy are foundational principles. A content firewall redacts sensitive information at both ingest and context-packing stages. Credentials and secrets never leave the local machine; a secure sandbox interfaces with cloud LLMs using reference IDs only. Furthermore, every sub-agent's access to memory is gated by a fine-grained capability token, ensuring least-privilege access. The Rust-based runtime is highly performant and can even forge new tools on the fly to solve novel problems. Symbiotic is designed for flexibility, allowing users to self-host on their own hardware and connect to any LLM provider, from local models via Ollama to cloud services like GPT, Claude, and Gemini.

Key features:

• Persistent AI Memory: A continuous loop captures unstructured data and distills it into a typed, encrypted 'Archive', creating a long-term, queryable knowledge base for AI agents.
• Collaborative Agent Swarm: Specialist agents reason over the shared memory, communicating by archiving findings as structured data, enabling complex, auditable, multi-agent collaboration.
• Zero-Trust Security: Features a content firewall for PII redaction, a local sandbox that isolates credentials from cloud LLMs, and capability-based tokens for granular agent permissions.
• Dynamic Tool Forging: The Rust-based runtime can autonomously generate, compile, and deploy new tools on the fly when encountering novel problems, extending its capabilities without manual intervention.
• LLM Agnostic & Self-Hostable: Run the entire system on your own hardware for maximum privacy and connect to any major LLM provider, including local models via Ollama or cloud APIs like GPT and Claude.
• Unified Recall Gateway: Intelligently packs context for agents by drawing from semantic, episodic, and procedural memory spaces, ensuring relevant information is always available for reasoning.