name: cognee repo: https://github.com/topoteretes/cognee-integrations subdir: integrations/hermes-agent sha: b132740960181e712f15c76ee80519cbd15fdb5d version: "1.4.1" description: "Knowledge-graph memory for Hermes, shared with Cognee's Claude Code, Codex and OpenClaw plugins: automatic recall, remember, forget, recaps and code search, on Cognee Cloud or a local cognee server. Disclosure — in local mode it installs cognee==1.6.5 and its dependencies (about 1 GB from PyPI) into a private environment, ~/.cognee-plugin/hermes/runtime, when you choose local mode (`hermes memory setup cognee` or `hermes cognee install-runtime`; never at session start), and runs a local cognee server from it (uvicorn on 127.0.0.1:8011); a session-close watcher runs in the background; each turn plus its tool calls and results (redacted, capped) go to that server and on to your configured LLM provider (OpenAI by default), or to Cognee Cloud / COGNEE_BASE_URL in remote mode; turns the server could not take are queued in ~/.cognee-plugin; the store lives in ~/.cognee, outside HERMES_HOME and shared across profiles and with Cognee's Claude Code, Codex and OpenClaw plugins; cognee sends anonymous telemetry unless TELEMETRY_DISABLED=1; the system prompt tells the agent to prefer Cognee memory over MEMORY.md/USER.md (COGNEE_MEMORY_STEER=false turns it off), and an answer that used recalled memory may end with a one-line Cognee credit, drawn with the Cognee logo by a transform_llm_output hook (COGNEE_PLUGIN_CITE_MEMORY=off turns it off)." maintainer: topoteretes tier: community category: memory docs_url: https://github.com/topoteretes/cognee-integrations/tree/main/integrations/hermes-agent#readme image: https://raw.githubusercontent.com/topoteretes/cognee-integrations/b132740960181e712f15c76ee80519cbd15fdb5d/integrations/hermes-agent/docs/banner.png readme: true capabilities: provides_tools: - cognee_recall - cognee_remember - cognee_forget - cognee_switch_dataset - cognee_code_search - cognee_recap provides_hooks: - on_session_end - on_pre_compress - on_memory_write - transform_llm_output provides_middleware: [] requires_env: []