framerslab/agentos

What it solves

AgentOS is a TypeScript framework designed to create AI agents that can maintain long-term memory, adapt their behavior based on personality traits, and autonomously create new tools to solve problems they cannot handle with their existing toolkit.

How it works

The framework combines several core systems to create emergent behavior:

• Cognitive Memory: Implements eight neuroscience-backed mechanisms (such as Ebbinghaus decay and reconsolidation) to manage how agents remember and forget information over long sessions.
• Runtime Tool Forging: When an agent encounters a gap in its capabilities, it writes a TypeScript function with a Zod schema. This code is reviewed by a separate LLM judge and, if approved, executed in a hardened node:vm sandbox before being added to the agent's permanent tool catalog.
• HEXACO Personality: An optional vector that biases the agent's retrieval, routing, and decision-making processes, ensuring consistent personality-driven behavior that persists even under context pressure.
• Orchestration: Supports multi-agent teams using six different coordination strategies (e.g., sequential, parallel, debate, graph) and provides APIs for DAGs, cycles, and goal-driven planning.

Who it’s for

Developers building complex, long-running AI agent applications that require persistent identity, sophisticated memory management, and the ability to evolve their own capabilities at runtime.

Highlights

• High Memory Performance: Achieves top open-source benchmarks on LongMemEval-S (85.6%) and LongMemEval-M (70.2%).
• Broad Provider Support: Unified dispatch interface for 11 LLM providers, including OpenAI, Anthropic, Gemini, and local options like Ollama.
• Soul Files: Allows agent identity and memory to be stored as a markdown wiki, making the agent's long-term memory human-readable and editable.
• Integrated Ecosystem: Includes built-in guardrails, voice/telephony pipelines, and 37 platform adapters for channels like Discord and Slack.