Claude Code is now multiplayer. Run `/delta share` in Claude Code, and your teammate joins in their own session with your full context and uncommitted work. You're both working on the same thread, backed by DeltaDB.

We're also launching the Delta CLI, so you can add DeltaDB support to your favorite tools. We'd love your help building plugins. Next up from us: DeltaDB in Zed and Pi.

pair programming finally beats teh merge conflicts

finally, agent sessions two people can actually share

The live pairing bit I'm less sold on. Handing a session over is the real win: picking up where a teammate stopped, with the dead ends they already ruled out, instead of rereading a diff and repeating them.

Exactly! It benefits both sync and async equally, whatever your preferred workflow is.

Sharing uncommitted work is the killer part. Humans and agents in one session is wild

shared context plus uncommitted work is the interesting bit. curious how it handles two people steering the agent in different directions

this is good stuff

I love Zed, I really do, but I just can't see the use case for this Live pair programming was added to VS Code ages ago with 24M downloads and a paltry 160 ratings and nobody I've worked with uses it. Doesn't everyone prefer to collab asynchronously at the PR stage?

You don't have to use sync collab, you can use it for handoff, and when our global provenance feature lands, your agent will be able to understand the decisions you made that led to the current code, as we will store that context, and you current agent can make smarter choices on the existing code.

what about i use codex and my teammate use claude ? do u support cross-harness-work? what if my feature need advice from sales or growth team or any other non-coding worker ? they have more context from customers, their opinons matter a lot more, and sometimes and they just use chatgpt web.

Any tool that can speak to the Delta CLI will be able to communicate with each other via a shared thread. We're working to add support for more harnesses / tools, but the community can jump in too.

does it handle both of you touching the same file or is that on you to sort out?

People and agents can be touching the same file. Delta was built for this use case. We know who edited what, down to the character. CRDTs ensure were conflict free.

Each person keeps their own Claude session, so the rules that load are the joiner's. If my user-level CLAUDE.md says never touch migrations, their agent picks up my half-done work without that line. Does the share carry rules too, or only the conversation?

Which ai subscription will be used here? mine or teammate?

The subscription of whoever is submitting to the agent. When I submit, it hits my sub, when you submit, it hits yours.

Would only use this read only at first. Two people steering one agent on uncommitted work sounds like a merge conflict happening live.

Shared context is the part solo setups miss. One note after this week's Downloads story: if a teammate's session pulls in your full context, it pulls in your CLAUDE.md too, so review instruction files like code

pair programming with an agent finally feels real

how does it even handle both of us prompting at the same time?

going to try /delta share on a debugging session with a teammate. curious how it handles two people prompting the same thread at once, does it queue or do you get interleaved edits?

shared context is half of review. the other half is your teammate asking the session why, on the line.

Your uncommitted changes now have witnesses

finally my teammate gets to watch me argue with claude live

sharing uncommitted work with full context is the clever bit. most pair-on-ai-work setups make you commit and push first, which is exactly when you don't want to. catching a teammate mid-mess is where the actual help happens

When a teammate joins, whose tool permissions and MCP credentials does the session run with? If it stays on the sharer's, I'd want to be careful what I hand over.

Handing over the reasoning, not just the diff. Now we're talking. 🚀

please tell us there's a co-op mode with friendly fire switched off

my teammate seeing my uncommitted mess live... no tidying up before push 😁

Shared context is the part that usually gets lost in handoffs. Curious how it handles two people steering the agent at once. Does one session lead, or do both prompts land in the same thread?

if "full context" means everything Claude read in the session, a .env it opened while debugging comes along too. does Delta strip secrets before the teammate joins, or is that on whoever runs /delta share?

Shared sessions with uncommitted work is the interesting part. How does it handle two people editing the same file at once?

Two writers still need a turn order

You can work in subthreads if you and your teammates want to work on the same thing, and let the agent merge the changes from each up to the parent. Or, you can both be commenting on the same thing in a comment change. Collisions don't really happen in Delta.

huge.

very exciting! any chance you'll support elixir syntax highlighting in the app soon? last big blocker for my team! thanks!!

Does delta support Claude now? It didn’t when I checked last time :)

Codex soon? @zeddotdev

Now two people can watch Claude write the code together. Pair programming, minus the programming.