Choose runtimes
Pick the agent runtimes that should share the same Lore memory graph.
Choose the runtime integrations that match your daily tools.
Runtime choices
| Runtime | Use when | Integration behavior |
|---|---|---|
| Claude Code | you use Claude Code for coding sessions | MCP tools, SessionStart boot injection, per-prompt recall, and CLAUDE.md guidance import |
| Codex | you use Codex CLI/Desktop | MCP tools, Lore skill, plugin registration, hooks feature, and compatibility user-level hooks |
| Pi | you use Pi agent | extension hooks inject startup guidance and per-prompt recall |
| OpenClaw | you use OpenClaw | plugin tools, startup health check, boot guidance, and recall injection |
| Hermes | you use Hermes Agent | Python skill/provider with boot, recall, and memory tools |
OpenCode (prerelease v1.3.15-pre.4) | you use OpenCode 1.18.3 or newer | native exact lore_* tools, Boot through experimental.chat.system.transform, and prompt Recall through chat.message |
Install only what you use:
Install only the OpenCode prerelease:
It installs lore-opencode.zip to ~/.config/opencode/plugins/lore-memory.js and shares ~/.lore/config.json with the other clients. The native tool list runs from lore_guidance through lore_move_node. System injection is fail-open, and prompt Recall continues when only the experimental system hook is unavailable. Standard OpenCode installation does not configure MCP; /api/mcp is a manual fallback only.
Every channel should use the same server URL and token. MCP-based clients include a client_type query parameter. Native OpenCode REST calls identify themselves as opencode without adding an MCP configuration.