Why does Claude Code forget my project every session?
Every Claude Code session starts with an empty context. It reads CLAUDE.md and its own memory files, and that's about it. Clyde adds a recall at the start of each task: its hooks make the agent search your notes before it acts, so last week's decision comes back without you retyping it.
What's going on
When a session ends, what the agent learned ends with it, unless someone wrote it into a file. Most of the time nobody did, so the next session asks how to deploy the repo again.
What Clyde changes
- One command connects Clyde and installs the hooks:
curl -fsSL https://heyclyde.me/install | sh - On a new task, the hooks refuse every other tool until the agent calls
recall. Quick follow-ups like "push it" don't trigger one. - What the agent figures out, it saves with
remember, so the next session gets it.
The same command sets up Codex, Gemini CLI and Antigravity, so they share what Claude Code learns.
Next
The full walkthrough is setting up the coding agents. If your CLAUDE.md has turned into a novel, read why long CLAUDE.md rules get ignored.
Questions
Does Clyde replace CLAUDE.md?
It replaces Claude Code's own memory files. The installer turns those off so there's one memory. Keep CLAUDE.md for the repo's instructions.
Will it slow every prompt down?
Recall runs on a new task, a folder switch, a 20-minute gap or a topic change. Quick follow-ups skip it.
Does it work in the desktop app's Code tab?
Yes. It uses the same Claude Code settings and hooks.