One user linked Hermes and OpenClaw through shared memory

A user split work between two agents — OpenClaw for research and Hermes for terminal-heavy implementation — and wanted them to share context instead of copying it back and forth by hand. They found a memory layer called cognee that let both agents connect to the same dataset, avoiding the need to build a custom bridge.

The actual workflow went: OpenClaw researches, Hermes implements, the outcome gets recorded, reviewer feedback is added, and future recall improves as a result. For example, OpenClaw researched CalDAV versus Google Calendar and recorded a preference for avoiding vendor lock-in; Hermes, starting in a separate session, recalled that context and used it to make an implementation decision, and OpenClaw was later able to recall Hermes's decision too.

The setup was also tested for whether memory could actually improve rather than just pile up transcripts. Before feedback, the guidance was to implement the Google Calendar API first as the fastest path — but that approach caused problems because Google's event types, OAuth scopes, provider IDs, and sync behavior leaked into the domain and created downstream issues.