Do Claude Code Child Sessions Share Memory?

Hey hey hey, welcome to Agentic Pod! I'm Liam.

And I'm Stella! Okay, so Claude Code just shipped a desktop update, and one parent session can now spin up child sessions. Like, um, a whole team of Claudes.

Yeah, I saw. Three run modes, local, cloud and SSH, and every child reports back to the parent when it's done.

So the real question is... what does a child session actually know when it starts?

That's the whole episode, so keep your seatbelt on.

Okay, so what is a child session, exactly?

So you ask Claude Code to split a big job, and the parent hands each piece to a child session. Each child works on its assigned task, and then reports the result back to the parent.

Mhm.

And the update adds saved environments, which can hold several folders, an SSH host or a cloud target, plus a worktree per folder.

And a little dashboard, right? In the sidebar?

Yeah, it shows which sessions are running and which ones are waiting for you. Plus a slash btw command for side questions, which, by the way, is a great name.

It's very "sorry to interrupt, but."

Exactly that energy.

So it's like hiring four interns in one click.

Four interns who each got exactly one sticky note. And that's where this gets interesting.

Wait, so what's on the sticky note?

Whatever the parent writes into the task. That's all a child starts with. Your decision from Tuesday, the naming rule you set last week, the reason you dropped that library? Not on the note, unless you typed it again.

And I'm the one typing it again. Four times. Because four children.

Right, and it gets worse. Every active session also consumes model usage, so re-explaining everything four times is not free.

Okay, but I use Cursor and Codex too. Their context is a totally different pile.

Yeah, that's the same problem wearing a different hat. Context lives inside one app, one session, one folder.

So the question is, how do the interns share a brain?

Okay, so, um, how does Vilix AI fix this? Technically.

Vilix AI is one memory layer that every session reads and writes over MCP, so it's the same memory for the parent, every child, Cursor, Codex, all of it. The task says "start by calling get context", and Vilix AI hands the child your decisions, your rules, what's already been tried.

So the sticky note just says "go read the notebook."

Exactly. And when the child finishes, it calls save turn, and Vilix AI stores the full exchange with a source label, so you can see which child, in which tool, decided what.

And when a child hits a wall?

Failures get saved too, with the exact error. So the next child doesn't walk into the same wall, and the parent can search by source to see what each one did.

What if I change my mind halfway through?

You say it once. In Vilix AI the newest memory wins, so every child that starts after that already has the new decision.

Okay, that's actually the part I wanted.

And here's the bit people miss. The same memory can also hold the plan for the next session, not just this one. Which is the question I want to leave you with.

Ooh, a cliffhanger. Okay, go.

What does the parent forget when you close the app? Cloud mode keeps sessions running after you close it, but your decisions still need a home that outlasts any one session. That's Vilix AI.

And the standing rules, like, always use pnpm, never touch the billing folder?

User rules hold how you work, project rules hold the facts for that repo, and skills hold the procedure, so every child loads the same playbook. And to steer a running agent, you drop a note in its Vilix AI agent inbox.

And for agencies, this one's perfect.

Agencies give each client their own Vilix AI account. Every parent and child session built for that client uses that client's API key, so the memory belongs to the client's project, and nobody builds memory from scratch.

Okay, recap time.

So, do Claude Code child sessions share memory? Each child starts with the task it was handed. Give every session one shared memory, get context at the start, save turn at the end, in Vilix AI. The setup guide, the article and the Vilix AI link are in the description.

And tell us in the comments how many children you'd run at once. I'm thinking four, and a snack.

See you next time on Agentic Pod.