Companies in the category 'Agent Memory'
These are companies that provide open source tools for managing and storing the long-term and short-term memory of AI agents, enabling them to learn and adapt over time.
AI agent telemetry and runtime infra
Paper Compute Co. builds open-source infrastructure for AI agent teams, providing tools to capture, inspect, and replay agent sessions. Its primary products are tapes, a zero-instrumentation telemetry layer that records every request and response between agents and model providers, and stereOS, a hardened NixOS-based operating system purpose-built for running AI agents in secure, sandboxed environments. The company also offers Paper Cloud, a managed platform for teams to collect, search, and learn from agent sessions at scale.
Open-source context database for AI agents
OpenViking is an open-source context database developed by ByteDance's Volcengine Viking team, designed specifically for AI agents. It unifies the management of agent context — memory, resources, and skills — through a file system paradigm, enabling hierarchical context loading and self-evolving memory. OpenViking supports hybrid vector retrieval, tiered context delivery, and observable retrieval trajectories to help developers build scalable, long-running AI agents.
AI memory framework for intelligent companions
NevaMind AI is the creator of MemU, an open-source memory framework designed specifically for AI companions that enables agents to truly remember, learn from experience, and deliver intelligent support that grows with users. The company specializes in providing memory infrastructure for AI applications, achieving 92% accuracy on memory-intensive reasoning tasks while delivering up to 90% cost savings through optimized retrieval and storage mechanisms.
Open-source memory OS for AI agents
EverMind is an AI infrastructure company based in San Mateo, California, focused on solving the long-term memory problem for AI agents. Its flagship open-source product, EverMemOS, is a Memory Operating System featuring a four-layer architecture—agentic, memory, index, and API/MCP interface layers—inspired by the hierarchical organization of the human brain. EverMemOS has achieved state-of-the-art benchmark results on LoCoMo (93.05%) and LongMemEval-S (83.00%), outperforming competing memory systems. EverMind is backed by Shanda Group, the investment firm founded by Chinese entrepreneur and philanthropist Tianqiao Chen.
COSS Weekly Newsletter
Stay up to date with the latest news, funding rounds, and announcements from the COSS universe.
Check out COSS Weekly on the web
