Coworker AI launches OM2 memory layer

OM2 targets the biggest enterprise-agent bottleneck: repeatedly rebuilding context before every task. The architecture is compelling, but its performance claims remain vendor-reported and should be validated on independent workloads.

• –A persistent neural graph could make cross-tool agents faster and more reliable than ad hoc connector chains or conventional document-level RAG.
• –Automatic stale-fact retirement and permission-aware traversal address two difficult production problems: outdated knowledge and data leakage.
• –MCP and API support make OM2 portable across Claude, ChatGPT, Gemini, Perplexity, and custom agents.
• –The benchmark used 114 customer-derived tasks across seven categories, but independent evaluations are still needed to substantiate the headline savings.
• –Pairing memory with model routing positions Coworker as an enterprise AI infrastructure layer rather than another standalone assistant.