framerslab/agentos
TypeScript AI agent framework: cognitive memory, runtime tool forging, multi-agent orchestration, 11 LLM providers.
What it solves
AgentOS provides a TypeScript framework for building AI agents that can maintain long-term memory, adapt their capabilities at runtime, and collaborate in teams. It addresses the limitations of stateless agents by implementing neuroscience-backed cognitive memory and allowing agents to programmatically create their own tools when existing ones are insufficient.
How it works
The framework operates through several core components:
- Cognitive Memory: Uses eight neuroscience-inspired mechanisms (such as Ebbinghaus decay and reconsolidation) to manage how agents remember and forget information over time.
- Runtime Tool Forging: When an agent encounters a task it cannot perform, it writes a TypeScript function with a Zod schema. This code is reviewed by an LLM judge and executed in a hardened
node:vmsandbox before being added to the agent's permanent tool catalog. - HEXACO Personality: An optional personality vector that biases how the agent retrieves information, routes tasks to specialists, and makes decisions.
- Orchestration: Supports multiple coordination strategies including DAGs, cycles, and goal-driven planning (missions) to manage multi-agent teams.
- Soul Files: Allows agent identity and long-term memory to be stored as markdown files (a wiki), making the agent's knowledge base human-readable and editable.
Who it’s for
Developers building complex, long-running AI agent systems that require persistent identity, high-performance memory retrieval, and the ability to evolve their own toolsets without manual developer intervention.
Highlights
- High Memory Performance: Top open-source benchmarks on LongMemEval-S (85.6%) and LongMemEval-M (70.2%).
- Extensive Provider Support: Unified interface for 11 LLM providers, including OpenAI, Anthropic, Gemini, and local options like Ollama.
- Multi-Agent Orchestration: Six distinct coordination strategies (sequential, parallel, debate, review-loop, hierarchical, graph).
- Integrated Guardrails: Five security tiers and six specialized guardrail packs for PII, code safety, and content policy.
- Voice and Telephony: Built-in pipelines for STT/TTS and integration with providers like Twilio and Deepgram.