OXO-B AGENTS
THE FLOW
From one conversation to merged work.
Talk to MPE
A small change? MPE just does it. A bigger feature? It calls APM in.
Plan together
You and APM shape the plan, and it shows up on the PLAN-TREE.
Cards and AUTO-FLOW
MPE turns the plan into task cards in dependency order. With AUTO-FLOW on, each ready card goes to a worker by itself.
Checked, then merged
A second agent checks the work and re-runs the tests itself. oxob commits and merges.
PARALLEL
Agents side by side, without trampling each other.
Each task works in its own copy of the repository. When two tasks write the same lines, oxob doesn't guess: it hands the conflict back to the agent that wrote the work.
- Workers never run git. They call one tool when they're done, and oxob does the rest.
- A card that depends on another waits for it, then starts by itself.
- In our tests, QA caught a worker claiming "passed" for tests it never ran.
IN THE WINDOW
Everything in one place.
One window per Mac, every project in a sidebar, each agent's own terminal right where you can see it.
The plan you can see
Plans, their parts and their cards, with what waits for what.
A model per complexity
Each card has a complexity level. Pick a basic model for simple cards and a stronger one for hard ones.
Work while you're away
Ready cards go to workers by themselves, whether you're there or not.
A second pair of eyes
Another agent checks each card's work and runs the tests itself, then sends it back if it falls short.
What happened, kept
Finished cards land in COMPLETED, with each session's story.
Notes between agents
Any agent can leave a report for you and for the other agents.
BRING YOUR OWN CLI
Your agents, your subscriptions.
oxob runs the CLIs you already use, signed in with your own accounts. No markup on tokens, and no telemetry: oxob sends nothing about your work to us.
Join beta 1.
Free and invite-only, for developers who already use Claude Code, Codex or agy every day. Write to us or say hello on Discord, and we'll send you the link.