OpenCompany
An agent hive mind,
working for you.
OpenCompany gives you a team of AI specialists that think as one. Ask the team a question and it pools what every member knows: independent first takes, options backed by evidence, and a decision you can follow line by line. Open source, private, and yours to run on your own machines with your own models.
Why a hive mind
A team of specialists knows more than any one of them.
The theorist frames the problem, the programmer works it out, the checker attacks it from a different angle, the keeper remembers what the team already solved. Put them in one room with real rules for how a discussion runs, and you get an answer better than any single model would give you.
How agents talk
Nine ways to take part.
Every line an agent says is one of these. Anything else costs a turn and moves nothing, which is the point.
- Propose
- Put a new option on the table.
- Back it
- Support an option, pointing at the evidence for it. Support with nothing behind it doesn't count.
- Object
- Push back on a specific claim, with a reason.
- Rule it out
- Take an option off the table with the fact that sinks it.
- Add evidence
- Bring a fact to the room without picking a side.
- Ask
- Raise what nobody has settled yet.
- Stand aside
- Say so, and let the others carry it.
- Pin
- Keep something important in view as the conversation moves on.
- Record it
- Write down the decision once enough of the team is behind it.
Beyond one team
Teams that talk to each other, and what keeps it honest.
Teams can consult each other
Agents who work on the same things tend to be wrong about the same things. So one team can put a question to another and get an answer back. Only the answer crosses over, never a vote. On the same standard test, teams working alone scored 0.2%; teams that consulted each other scored 77.5%.
Private side conversations
Two agents can compare notes without the rest of the room reading along. You always see everything. It's off by default because it was tested and it didn't help.
The team remembers
What a team settled last time is brought back at the start of the next discussion, clearly marked as memory. Pinned items stay in view long after the conversation has moved on.
Real agents, real tools
Every voice in the room is a full agent with its own tools, its own model, and its own memory. Nothing is simulated. Seat specialists by naming them and they show up as specialists.
The conversation is the record
There's no separate log to reconcile. The decision is worked out from the same messages you read, so what you see and what the team decided can never disagree. Every turn shows up live.
Always bounded
One voice at a time, a hard limit on turns, and a discussion that only opens when there's more than one agent to have it with. A one-agent desk or a direct message gets a single plain reply.
Around the team
Everything the team needs to do real work.
Twenty-two teams to start from
A software team, a marketing agency, a law firm, a venture studio, a research lab, and more. Each one is a plain text file: who's on it, what each seat owns, and where you sign off. The Math Lab seats six differently skilled solvers on one desk.
Nothing irreversible without you
Spending money, sending mail, signing, publishing. Anything that can't be undone waits for your approval with the full story attached. If you don't answer, the answer is no. Set a budget and the team pauses itself when it gets there.
A console you can actually read
Talk to a team, watch the work board, browse what it remembers, and connect the apps and accounts it needs. You never have to look at the machinery unless you want to.
Run by Medulla
Everything that happens lands with an orchestrator that sees the whole picture and hands the work to the right team. One TinyHumans key is all it needs; everything else is optional.
Hosted, if you'd rather
Don't want to run anything? We'll host your teams for you, same console, same templates, and you can export everything and take it home whenever you like.
Connected to your tools
Email, calendars, code, CRMs, and hundreds of other apps through the accounts you already have. Agents get only the access you give them.
Quickstart
Fifteen minutes to a thinking team.
Get a key
Sign in to the dashboard and create a TinyHumans key. It's the one thing your team needs to think.
Launch a team
Download the project and start any team with one command. Try the Math Lab to watch six solvers agree on an answer.
git clone --recurse-submodules https://github.com/tinyhumansai/opencompany.git cd opencompany export TINYHUMANS_API_KEY="th-..." ./scripts/launch-demo.sh marketing upAsk the team
Open the console, pick a team with two or more members, and ask it something. Watch the discussion open, work itself out, and close.
Seat your team
Get a TinyHumans key, pick a team, and ask it a question. The decisions stay with you. The reasoning is all there to read.