codemem is designed to sit inside AI coding-agent workflows, capture useful context automatically, and surface it again when the next session needs it.
OpenCode
OpenCode is a first-class fit. Run npx -y codemem setup --opencode-only, restart OpenCode, and let the
plugin manage backend execution automatically. The setup command also registers the explicit memory tools.
Claude Code
Claude Code is also a first-class surface. Add the codemem project's own marketplace source, install the plugin, then reload plugins. This marketplace belongs to the codemem project and does not imply official Claude approval. Current builds support automatic viewer startup and a sidecar runtime path for subscription users without a separate API key.
Codex
Codex is supported through its own plugin marketplace (codex plugin marketplace add /
codex plugin add codemem@codemem). The plugin captures session activity through Codex hooks into the same
shared pipeline as OpenCode and Claude Code, and automatically injects relevant memories into your prompts at submit
time via the UserPromptSubmit hook. Bundled MCP tools add on-demand recall when you want to search memory
explicitly. Validated targets are Codex CLI 0.135+ and current Desktop builds.
API-key Codex Desktop: If plugin installation is unavailable for a non-subscription or API-key Desktop
setup, run npx -y codemem setup --codex-only. Restart Codex and approve the one-time prompt to trust the
codemem hooks. MCP recall works immediately; hook commands use a global codemem when present and otherwise
use an npx launcher.
What the integration model is for
The point is not just to save transcripts. The point is to keep decisions, debugging context, and repo-specific patterns available when a later agent session needs them.
More to come
OpenCode, Claude Code, and Codex are supported integrations. More agent surfaces can follow from the same model.