Frequently asked

No hedging. No fine print.

Before @Cohenta joins your Slack: what installs, where your keys live, what Cohenta receives, who holds the last gate.

The question wall12 questions · 04 topics
in plain languageanswer path
?you askyour factsflag + annotatenever silently rewrittenhuman gategrounded answer
Every answer takes this pathflag + annotate · human gate

In plain language

The short of it.

Install & connection

03 questions
What do I install?
Two things: the Runner — a small engine on your laptop, server, container, or cloud — and your own Slack app, created from Cohenta's manifest. Slack issues the tokens to you; they land in your keychain, never at Cohenta.
Which tools can I use it from?
Seven surfaces: Slack, Microsoft Teams, ChatGPT, Claude Code, claude.ai, the Anthropic API, and a CLI + API. A campaign started in one can be checked from any other.
Do I need a public endpoint?
No. The Runner connects outbound only — Slack over Socket Mode, an outbound WebSocket; ChatGPT through OpenAI's secure tunnel. No inbound port, no public endpoint, nothing to expose. It works behind a firewall, even from a laptop.

Keys & custody

03 questions
Where do my keys live?
In your keychain or vault — macOS Keychain on a laptop; Azure Key Vault, AWS Secrets Manager, or HashiCorp Vault in the cloud. Config stores references, not values; the Runner resolves them locally at call time.
What does Cohenta receive?
Cohenta never receives or stores your provider keys, publishing credentials, canonical assets, or campaign state. Data is shared only with the platforms and providers you have chosen — Cohenta's cloud is never in the execution path.
Who owns the publishing connections?
You do. Every channel uses your accounts and your credentials, connected in your environment and stored as references — never uploaded. Disconnect a channel and the access was always yours to revoke; Cohenta has nothing to hand back.

Approvals & grounding

03 questions
How do approvals work?
The run pauses at the gate and waits — durably, across restarts. Approve, Request changes, and Reject buttons in Slack or Teams resume that exact paused run; publication takes a separate explicit confirm, and ad creative lands paused.
How does Cohenta stop fabrication?
Every statistic and multi-digit number is validated against the facts and KPIs you provide. Anything that does not trace to a source is flagged and annotated — never silently fixed or blocked — and reaches a human before it ships.
Can I trace where an asset came from?
Yes. Every run writes a per-asset manifest on your Runner — what passed, what was flagged, which engine rendered it, which sources it traces to — stored with your campaign state, not with Cohenta.

Output & pricing

03 questions
What can it actually produce?
Posts, articles, visuals, short video, decks, and landing pages — every asset the campaign needs, with human approval on the way out. Much of it renders deterministically: pixel-identical PDF and editable PPTX from one spec.
What is free?
Cohenta Skills — free in ChatGPT and Claude. Strategy, brand-voice assessment, repurposing, and copy review, reasoning-only: no Runner, no backend, no keys. When you want campaigns executed, the Pro Runner picks up where Skills stops.
What does the license cover?
License verification, downloads and signed updates, documentation, and support — priority on Team Deployment. The engine executes on your hardware either way; the license keeps it verified, current, and supported.

Under the hood

Enforced in code, not promised in a prompt.

Gate verdict, grounding trail, engine, lineage — every asset ships its manifest, written on your Runner.

Next step

Still have questions?

Start free with Cohenta Skills — or go deeper on custody and what Cohenta never holds.

Start freeHow custody works