Hi PH 👋 I build with Claude Code, Codex and Grok side by side, and I kept noticing the same thing: session ten was not smarter than session one, just more expensive. Every new thread started blank and I re-pasted the same context forever. So Solenta injects a local memory server into every agent session automatically. What one thread learns, the next one knows across all six CLIs. SQLite on disk, localhost only, bearer-token gated. There's a hypothesis ledger so the next agent inherits what the last one already ruled out, and file:line citations get checked against your worktree before a fact is injected, so stale memories get invalidated instead of poisoning the next session. The rest is the desk you need to run those agents: git as a real pane, worktrees, a terminal, an embedded browser, and a planboard that is just your GitHub issues. I'd love to hear: which two agents do you actually switch between, and what context do you find yourself re-explaining every time? That's the thing I'm building for.

Solenta

Shared memory for Claude Code, Codex, Cursor & Grok

Shared memory for Claude Code, Codex, Cursor & Grok