Connect agents
Connect Codex, Claude Code, Pi, OpenClaw, Hermes, or any MCP client to the same Lore memory graph.
Connect agents after the server is healthy and /setup is complete.
Supported runtimes
| Runtime | Integration | What agents get |
|---|---|---|
| Claude Code | MCP tools, session-start hooks, CLAUDE.md guidance | Boot context, per-prompt recall, memory read/write tools |
| Codex | MCP tools, plugin hooks, skill registration | Boot context, per-prompt recall, memory management |
| Pi | Extension hooks | Startup guidance, per-prompt recall |
| OpenClaw | Plugin tools | Boot guidance, recall injection, health check |
| Hermes | Python skill/provider | Boot, recall, memory tools |
All channels share the same Lore server and memory graph. A rule saved through Claude Code is visible to Codex, Pi, OpenClaw, and Hermes.
Connection paths
Supported installer channels are claudecode, codex, pi, openclaw, and hermes. You can install all channels, select a subset, skip Docker, or point local agents at a Lore server running elsewhere.
Agent integration has two different jobs:
- Connection — MCP URL,
client_type, bearer token, plugin files, and hooks. - Memory ownership — shared durable rules live in Lore, while local
CLAUDE.mdandAGENTS.mdfiles stay small.
This section covers connection behavior. After the connection works, use Move local agent docs into Lore to remove duplicated long-term instructions from local agent files.