SuperAPI AI Directory

kortix-voice

Design & creation

View repository

Updated 10-15 23:47

About this skill

Kortix Voice

How Kortix reads. The design system governs how it looks; this governs the words. The canonical positioning is suna/COMMS.md — when in doubt, that doc wins; this skill is how you apply it in the product.

The one rule

Plain language by default. Technical detail is opt-in.

A non-technical founder must understand every default surface without knowing what a repo, branch, sandbox, or commit is. The technical truth still exists — it lives one layer down: behind an "Advanced"/"For developers" disclosure, in an "Under the hood" callout, or in the Reference docs. Never lead with jargon; never hide the outcome behind it.

Voice

  • Outcome first. Say what the user gets, then (optionally) how. "Your agent drafts the report" > "Spawns a session that runs the prompt."
  • You, active, short. Address the user as "you". Active voice. Short sentences. Sentence case for everything (headings, buttons, labels) — not Title Case.
  • Calm and confident. Founder-grade, plain. No hype, no fear, no filler.
  • Specific, not salesy. Concrete nouns and verbs over adjectives.

Vocabulary (the command center)

Use: command center · agents · skills · integrations · connections · automations · sessions · projects · accounts · memory · deliverables · on-demand / human-assisted / automated · review · merge · open · own your infrastructure · 24/7 · compounds.

Avoid (banned):

  • Product mislabels: chatbot · copilot · "assistant" (for the product) · "AI employee" (singular).
  • Hype/filler: simply · just · seamless · powerful · blazing · effortless · revolutionary · "in seconds".
  • Fear: never scare the user into a click.
  • Specific dead phrases: "no Git account required" · "managed repo / sandbox snapshot / default branch" on a non-technical surface.
  • The legacy name: do not write "Suna" in any new copy. The product is Kortix.
  • Runtime pinning: the sandbox runtime is becoming pluggable — don't hard-pin or prominently brand it (e.g. "powered by OpenCode") in product/marketing copy.

License wording

Always say "open" / "source-available". Never "MIT", never "100% open source".

Plain ↔ technical glossary

Lead with the plain term; reveal the technical one only in Advanced/Under-the-hood/Reference.

Concept Plain (default UI) Technical (opt-in)
Project a workspace for one company, product, or idea a git repository + kortix.toml
Session one task you hand your agent, in its own safe space an isolated sandbox VM on a branch
Change request the agent's proposed work, for your review a branch merged into the default branch
Keep / approve "merge" / "keep the work" merge the change request
Secret a key your agent needs, stored safely encrypted env var injected at runtime
Connection connect a tool your agent can use [[connectors]] + the Executor
Trigger run a task on a schedule or event [[triggers]] (cron / webhook)
Account your account (personal or team) the tenant; roles owner/admin/member

Before → after

❌ Don't ✅ Do
"Start with a private managed repo. Existing GitHub repos can be imported." "A dedicated space for one company, product, or idea."
"Creates a private managed repo, seeds the starter, builds the first sandbox snapshot." "We set up your project, ready to use."
"Kortix-managed repository" "Start fresh (recommended)"
"Import existing GitHub repo" (prominent) tuck under "Already have code? (for developers)"
"Run agents in disposable VM sandboxes against your GitHub repos." "A workforce of AI agents that does the real work across your tools."
"Powered by OpenCode" (omit; the runtime is becoming pluggable)

Technical surfaces are different

Developer-facing surfaces — the Reference docs, the CLI, kortix.toml, code comments — should be precise and dense, not dumbed down. Plain-first applies to the product UI and the conceptual/marketing layer; the reference layer states exact fields, flags, and behavior. Accuracy always beats simplicity: never make a claim the code doesn't back (verify against source).

Checklist before shipping copy

  1. Could a non-technical founder understand it with zero jargon? If not, simplify or move the jargon behind a disclosure.
  2. Sentence case? Active voice? Outcome-first?
  3. No banned words? No "Suna"? No runtime pinning?
  4. Is every factual claim true to the code?
  5. Does it match COMMS.md positioning and the glossary above?