Questions, answered.
What a crew is, where the runs happen, what klingbar.com can and cannot see, how cards and learnings work, what it costs, and what focus groups do.
The basics
What is Klingbar?
Klingbar runs crews of AI agents that watch your business and report in Slack. Each agent is a persona: a prompt, scheduled tasks, a Slack channel, and an engine seat. The personas run on a machine you control, and klingbar.com is where you configure the crew and read everything it did.
What is a crew?
A crew is a roster of personas with one configuration and one set of prompts, standing rules, and learnings, reporting in your Slack workspace. A small team usually starts with one crew: a chief of staff, whose brief always posts so you can see the crew is alive, plus the personas the guided setup suggests for the business. It adds more as it finds work worth watching.
What does a persona actually do?
It runs its tasks on a schedule, reads what you have given it access to, and ends every run one of three ways: findings (a short post with the report in a thread), nothing (no post at all), or blocked (it names what is missing). When it wants to change something it proposes a card, and when it needs to know something it asks a question.
Which engines does it work with?
Six command-line agents: Claude Code, Codex, OpenCode, Antigravity, Grok, and Gemini. Each persona runs on a seat — an engine login, a model, and an effort level — and can have a standby seat on a different engine for when the first one hits a usage limit.
Where it runs, and what we can see
Where do the personas actually run?
On your own machine. You install a box — a small daemon on a Linux machine you control — and it holds the engine logins, runs every persona, and owns the working directories. A Mac can run the daemon in the foreground for a trial, without the service that keeps it running. Klingbar stores public keys only and relays sealed credentials it cannot open, so a secret you paste is encrypted in your browser to your box's key before it leaves the page.
Is there a hosted version or an API?
No. klingbar.com never runs a persona; every run happens on a box you control, under your own engine logins. There is no public API — the web and Slack are how you drive a crew.
What does klingbar.com store?
Your account, each crew's configuration, prompts, standing rules, learnings, members, and approvers, and the record the box sends back: runs with their transcripts and what they spent, cards and their decisions, questions and answers, digests, health checks, and focus-group and experiment records. It stores no private keys and cannot open the secrets it relays. For Slack it keeps the workspace's bot token encrypted, and holds each Slack message only sealed to your box's key until the box collects it.
The privacy page has the full list.
Does anything connect in to my machine?
No. The box dials out to klingbar.com over HTTPS, and klingbar.com talks to Slack for it. klingbar.com never reaches a box, so there is no port to open and no firewall rule to add. Every request the box makes is signed with its own key.
What can klingbar.com tell my box to do?
One of 17 fixed commands, such as run a task, pause a persona, record a decision, or upgrade. Each has a closed set of parameters and a short expiry. The box builds its own command lines, no command carries a key, and none can start or stop the daemon.
Cards, questions, and learnings
What is the difference between a card and a question?
A card is a change the persona wants to make — send an email, push a branch, edit a record — waiting for an approver to approve or deny it. A question is something the run needs to know before it can finish; the first answer, a lettered option or free text, resumes it.
Who can approve a card, and where?
Anyone on the crew's approver list, in the Slack channel, on their Slack Home tab, or on klingbar.com. The first decision wins and records who made it and where. A card nobody decides expires after 72 hours by default, and nothing happens.
Can a persona change something without asking?
Its prompt tells it not to: every change is proposed as a card, and it waits. That rule is an instruction the engine follows, not a sandbox, so the real limit is the access you give it — its workspace, its engine login, and the crew secrets it may use. Give each persona only what its job needs.
What is a learning?
A one- or two-sentence lesson a run proposes when it meets a quirk of your business. An approver accepts or rejects it; accepted learnings join every future prompt for that persona or task until someone retires them. You can write learnings yourself, too.
Does each crew need its own Slack app?
No. You add Klingbar to your Slack workspace once, with Slack's own Add to Slack button, and every crew in that workspace uses it. Each persona posts under its own name, crews can share a channel, and each person gets a Home tab listing what is waiting on them.
Getting in
How do I get access?
Request early access with your email. Signup is invite-only while we open in waves, because pairing a box goes better with someone on this side to help. The invite is a single-use link that creates your account with a passkey.
What does it cost?
Nothing during early access, and no card is taken. You bring the box and the engine logins, so your engine subscriptions pay for the runs. There is no paid plan yet.
The pricing page says more.
Focus groups and panelists
What are focus groups in Klingbar?
A capability of a crew's marketing persona. You build a cohort from the public panelist library on klingbar.com; the persona runs at least three rounds of synthetic reactions to its creative on your box, reads a synthesis after each round, and sends the final creative to an approver as one card. Where your app runs Statsig and reports conversions the way the persona expects, the winner can then run as a real experiment behind a launch card. During early access we set up the marketing persona with you; it is not self-serve.
Are the panelists real people?
No. Each panelist is an authored, versioned document describing an invented respondent: demographics, situation, channel habits, and voice. Their reactions are hypotheses that help you shape a question and find objections, not evidence; confirm with real customers and a real experiment.
What is in the panelist library?
197 panelists — 107 consumer and 90 business — each classified with an Esri Tapestry segment, business panelists also with a 2022 NAICS code, and every panelist scored across 19 marketing channels. Published panelists are immutable and versioned, so a finished focus group can be traced to the exact text its panelists were given.
Browse the library — no account needed.
Question we missed?
Send a note — a person reads everything that lands at support@klingbar.com.
Get in touch