your entire agent team is now one yaml file. roles, who reports to whom, what each one is told to do. write it once, version it, and bring the whole team up together instead of clicking through five separate agent setups.

---

what shipped

team-import v1, in the open-source 5dive cli (one bash file, MIT). a team isn’t a pile of agent create commands you typed once and forgot anymore. it’s a declarative file.

you do create the bots first: each role reads its own telegram token from the env, and the importer stops cold if one’s missing. so set your tokens, then:

5dive team import startup

one command brings the team up, wired together. or point it at your own file with 5dive up -f my-team.5dive.yaml. tear it down with 5dive down, check state with 5dive ps.

---

the file is the org

a role in the spec carries more than a name:

• its own model and effort. lead roles on opus/high, the workers on sonnet. set per agent, in the file.
• reports_to. this is the good part. it draws the org chart, and it writes the exact 5dive agent send delegate and escalate commands into each role’s instructions. the delegation lane, spelled out for every agent.
• a generated CLAUDE.md. each role gets its own standing instructions, with a role and reporting block baked in from the spec.
• seeded goals. the first work for each role lands in its task queue on import, waiting for it.

change the file, re-import, and the org changes. your company has a diff now.

---

two teams you can clone today

ships with two:

startup — a founding team. CEO, CMO, DevOps, a competitor researcher, and a creative. the CMO and DevOps report to the CEO, and the researcher and creative sit under the CMO. an actual reporting structure, not five agents in a trenchcoat.

content-studio — an editorial team. editor-in-chief over a writer, an SEO, a designer, and a distributor. note it’s not one role each: four people report to the editor, and nothing stops you adding three more writers if that’s the bottleneck. also note none of it is “point an agent at a repo.” it’s a content shop.

5dive team import content-studio --auth-profile=<account>

both come with the roles written out, so you’re cloning a team that already knows its job. the yaml for both lives in the repo if you want to read or fork it.

---

why a file beats clicking

because a team that lives in a file is a team you can diff, version, and hand to someone else. the two bundled templates are exactly that: someone wrote the org down, and now you stand it up on your own box. write your own and it’s yours to fork and improve, like any other code in your repo.

and 5dive export captures the shape of a live fleet back out to yaml: every role, its model, and who it reports to. the org chart you built by hand, readable as a file. (it captures the structure, not each role’s instructions, so read it as the blueprint, not the whole team.)

---

try it

run 5dive team ls to see the bundled teams. set your role tokens, then 5dive team import startup to bring one up. or 5dive export your current fleet and read your org as a file.

rolling out to managed boxes on the nightly update. self-hosted: 5dive self-update.

---

spin up a team at 5dive, or grab the open-source cli and write your own on a box you’ve already got.