Org-shaped agent swarms
Describe your business once. Mitosis drafts a CEO + departments + individual contributors with explicit responsibilities and KPIs. Re-org any time.
★
╱ ╲
★ ★
╱╲ ╱╲
★ ★ ★ ★Spin up an entire org of AI agents — a CEO, departments, ICs — that talk to each other, use your tools, and ship work while you watch the terminal scroll.
$ mitosis onboard > hi! tell me about your company in one sentence. < we're a 6-person dtc skincare brand. > got it. drafting an org… [ok] ceo agent + 3 departments + 7 ICs created. $ run "launch the spring campaign" [plan] marketing.head → 4 sub-tasks [tool] generate_image · brief → hero asset [hold] approval needed: send email blast (12.4k recipients) $ approve a3f1 [ship] objective complete in 7m 42s · 11 tool calls
The CEO agent spawns departments. Each department spawns individual contributors. Each IC picks up tools and gets moving. You watch the office come alive — like cells in mitosis, dividing on cue.
Describe your business once. Mitosis drafts a CEO + departments + individual contributors with explicit responsibilities and KPIs. Re-org any time.
Drop an objective. The CEO routes it to the right department, which spawns sub-agents. Watch the tree execute in real time, pause, redirect, or approve.
First-class integrations for email, calendars, CRM, payments, image generation, code, web search. Risky actions pause for human approval — never silently fire.
Workspaces, roles (owner / admin / member / viewer), per-tenant API keys, encrypted credentials, scoped audit log. Bring your own LLM keys or use ours.
Models from Replicate (Flux, SDXL, Imagen, etc.) wired into the agents. Marketing assets, mockups, hero images — generated inline as part of an objective.
Toast in-app the moment an objective ships or an approval is needed. Email digests via Brevo. Owners + admins only — no inbox spam.
Mitosis is opinionated about org structure so your agents always know who they work for, what they own, and what they may touch.
ceo (claude-3.5-sonnet) ├─ marketing.head │ ├─ copywriter │ ├─ designer · [tool] generate_image │ └─ analyst · [tool] ga4, mixpanel ├─ sales.head │ ├─ sdr · [tool] hubspot, gmail │ └─ ae · [tool] calendly, stripe ├─ ops.head │ ├─ recruiter · [tool] greenhouse │ └─ accountant · [tool] xero, plaid └─ engineering.head ├─ backend.eng · [tool] github, linear └─ frontend.eng · [tool] github, vercel
No. Mitosis encodes your org chart and routes objectives along it. Agents have explicit roles, scoped tools, and gated actions. The terminal is the control surface — every approval, audit, and credential is first-class.
Yes. Anthropic + OpenAI keys are stored encrypted per workspace. Self-hosted models work too if you point to an OpenAI-compatible endpoint.
Tool calls have a risk_level. Anything destructive (sending email, charging cards, posting publicly) requires an approval — surfaced as a toast + email to owners/admins.
Cloud: EU + US regions, Postgres + S3-compatible storage. Self-hosted: anywhere you run Kubernetes or Docker.
We ship a Docker Compose file for solo, and a Helm chart with KEDA-based autoscaling for production Kubernetes.
╔════════════════════════════════════════╗ ║ $ mitosis onboard ║ ╚════════════════════════════════════════╝
Give us five minutes. We'll hand you a fully-formed swarm ready to ship work tonight.