Hivemind · Agents
Agents that already know your company
A Hivemind agent does the work that repeats, at its own moment, and starts from what your company already knows. So the second agent begins where the first one left off, and none of them has to be taught your business again.
What an agent is
Four parts, all in your hands
An instruction
What it should do, in plain words. Write it yourself, or describe the agent to Claude or ChatGPT and let it draft one.
A toolbox
Only the tools you switch on, from the systems you connected. Every tool is marked read, write or irreversible.
A moment
Every hour, day, week or month. Or a set time before or after the meetings in your calendar.
A report
What it found and did, written into the knowledge base, with a message in Hivemind or by mail.
The foundation
Set up once. Every next agent starts ahead
Most AI projects begin by teaching the tool the business again. In Hivemind that work is kept: what one agent needed, the next one already has. And so do Claude, ChatGPT and Copilot, for everyone who asks.
- Definitions What a customer, an order or a project is to you.
- Decisions Who decides what, what was decided and why.
- Rules Prices, limits, the way of working, what needs a yes.
- Connections Your systems, connected once for every agent.
Hivemind · set up once, reused by all
Agents at work
The work that comes back every week
Six agents, each on systems Hivemind connects today. Each shows its toolbox: what it may read, what it may change, and what waits for a person.
Meeting preparation
two hours before every weekly meeting
Meeting preparation, two hours before every weekly meeting: Reads the invite, the decisions of the last meeting and what is still open, and posts an agenda with the points that need a decision.
3 allowed · 1 does not exist for this agent
Meeting debrief
thirty minutes after the meeting
Meeting debrief, thirty minutes after the meeting: Turns the notes into decisions and actions, writes the decisions where they belong and sends everyone their own list.
4 allowed
Bookkeeping
every weekday at 08:00
Bookkeeping, every weekday at 08:00: Finds the invoice or receipt behind each new bank line, processes new purchase invoices, and asks the person who knows about the rest.
2 allowed · 1 waits for a person · 1 does not exist for this agent
Inbox triage
every hour, during working hours
Inbox triage, every hour, during working hours: Labels new mail, drafts the answers the company already knows for a person to send, and flags what needs someone.
4 allowed
Campaign report
every Monday at 09:00
Campaign report, every Monday at 09:00: Reads last week's campaigns and flows, compares them with the targets in the knowledge base, and posts what stood out.
3 allowed · 1 does not exist for this agent
Knowledge upkeep
every Friday afternoon
Knowledge upkeep, every Friday afternoon: Finds broken links, stale dates and numbers that no longer match the file that owns them. Fixes those, and proposes the rest.
2 allowed · 1 waits for a person · 1 does not exist for this agent
More per line of work, from tour management to engineering firms: every solution.
Trust
Freedom it earns, one run at a time
A new agent proves itself before it works alone, and it never gets a tool you did not give it. The full idea: agents with limits.
Try it first
A test run shows what the agent would do and changes nothing.
Start supervised
You watch its first runs. When none of its proposals was rejected, it runs on its own.
Choose how far it goes
Only propose. Fix the mechanical things and propose the rest. Or write freely. Every change can be undone.
A yes for what matters
An irreversible action waits for a person, in Hivemind. The AI that asked can never approve it.
A limit on every run
A cap on cost and length. Near the limit it wraps up and still writes its report.
A record of every run
What it read, what it decided, what it changed and who approved it.
Learning
Better with every run
An agent keeps a notebook. Each run writes down what the next one should know: the supplier who always sends two invoices, the client who wants figures in thousands. You see every note on the agent's page, and remove what is wrong.
How one is made
- Described in your own AI tool. Tell Claude or ChatGPT what you need, and it drafts the agent in Hivemind.
- Checked before it is saved. Hivemind reads the draft and says what is unclear, risky or missing.
- With your own skills. An agent can work as one of your assistants: the same instructions and tools your people use in chat.
- Changed whenever you like. The instruction, the toolbox and the moment, by you or by the AI tool that made it.
Two ways in
Build them yourself, or with us
Make your own agents
Fill the knowledge base, connect your systems and switch on a first agent. It runs in the Dienox cloud, or in your own environment.
Have them built with you
We find the work worth automating, build the agents and the connectors they need, and stay on until they run on their own.
What is a Hivemind agent?
An agent that does recurring work for your company on its own: on a schedule, or before and after your meetings. It starts from what your company knows in Hivemind, uses only the tools you switched on, and writes a report of every run. What could do harm waits for a person.
How is it different from an agent in ChatGPT or Claude?
It does not need anyone to have a chat open: it runs at its moment, in Hivemind. It works from the knowledge of the whole company instead of one person's chat, has a toolbox with a level on every tool, and earns its freedom over supervised runs. Your people still use Claude or ChatGPT to make one.
Which systems can an agent use?
The connectors in Hivemind today: the knowledge base and its files, Gmail, Google Calendar, Google Drive, Slack, Notion, SharePoint, OneDrive, Moneybird and Klaviyo. For a system that is not on the list, we build the connector, and every agent can use it from then on.
Who builds the agents?
You can, in Hivemind: describe what you need to Claude or ChatGPT, and Hivemind checks the draft before it is saved. Or we build them with you: we find the work worth automating, build the agents, and stay on until they run on their own. That is Automate.
Get ahead with your first agent
Start with one piece of work that comes back every week. Sign up in two minutes, and we come back to you within two working days.