Every Trupe agent has a name, a role and a corner of its own to work in. Your team talks to them like any other colleague, and always knows what each one is doing.
Runs on your own server. Works with Claude Code and Codex.
Lia, Bento, Nina, Otto, Cora and Teo. Each team puts together its own crew, and each person can have one too.
Give it a name, pick the team and say what it's for. It takes a minute.
Ask in the chat, the way you'd ask someone at the next desk. It answers, asks when it's unsure and shows what it did.
Everything each agent does is recorded, and everyone on the team sees what they need to see.
All on the same team. Hover to meet them. Click to flip.Swipe sideways and tap to flip.
“I put the client's proposal together. It just needs your okay.”
FlipTalks to everyone and writes in the team's voice. Prepares proposals and waits for someone to check them before sending.
“I answered the simple ones. The three tricky ones are waiting for you.”
FlipCalm and patient. Nobody leaves a conversation with him without an answer, and he knows when to call in a person.
“Found the bug. I opened a small PR and wrote the test with it.”
FlipReads the whole codebase before touching a single line. Prefers small changes and explains the why of each one.
“Month closed. The margin dropped because of two late invoices.”
FlipReconciles accounts and spreadsheets and explains the number before you ask. Shows all the math and speaks up early when something doesn't add up.
“I summed up the meeting and sent the next steps.”
FlipComes back with sources, not hunches. Cites where each thing came from and says clearly when she isn't sure.
“I moved the dentist appointment and reminded you about the birthday.”
FlipLooks after your life outside work: schedule, household bills and reminders. Stays with you only, nobody on the team sees it.
The names are examples. In your Trupe, you choose.
Swipe to see all eight.
Each agent works in its own space, with its own things. What belongs to one never mixes with what belongs to another. When the work calls for it, you put a few agents in the same space.
Invite people from your company with a link. Each one comes in with a role: who looks after things, who creates, who follows along.
An agent only talks to another agent in the same group or with your permission. No side conversations nobody asked for.
A timeline shows what each agent did and when. It's kept, with no editing.
Company keys live in a vault, and not even the screen shows the value once it's saved.
Claude Code or Codex, with each person's subscription or with the company key.
An agreed limit per team, so nobody gets caught by surprise. When it's reached, the agent takes a break and waits for an admin.
Each agent will have a text that says who it is, written by the team, and a memory that grows with use.
By default, one agent doesn't talk to another. Cora can pass a meeting summary to Nina because they're both in the same group. Lia, from another team, only joins with permission. Every attempt is recorded. Direct conversation between agents in the chat is coming soon.
Trupe is a crew of agents that works for the whole team, under the company's rules. Anyone who wants to use it alone can, but the focus is the team. We asked three well-known agent products the same six questions. Their answers come from the documentation they publish themselves. Where they do the same, it says so. Where they do better, it says that too.
| Question | trupecompany agents, for the whole team to talk to | Hermes Agentby Nous Research, an open agent for each person | OpenClawopen assistant, on WhatsApp, Telegram and Slack | Grok Botby xAI, bots for each person and for the team |
|---|---|---|---|---|
| Do the agents belong to the company or to each person? | The companyThe agents belong to the company, not to someone. Each person on the team has a role that says what they can change. | Each personBuilt to learn who you are. A bot can be shared with the team on Telegram. | Whoever installs itA team that trusts each other can share. Anyone who needs separation uses another gateway. | Each personOne bot per person. The team bot is the exception, and it has an owner. |
| Does each agent have its own corner? | YesIts own folder and processes, on Linux with Landlock. One can't touch another's. | Only with DockerProfiles separate memory and keys, but don't isolate. Docker does the isolating. | If you turn it onOff by default. The sandbox can be turned on per agent. | Per personOne person's bots share the same computer. |
| Can the agent see the company key? | It can'tIt only holds a placeholder. The key is added along the way, outside its corner. | With the proxy onOff by default and only works with Docker. When on, the agent only sees a placeholder. | With the proxy onOff by default. When on, the agent only sees a placeholder. | Uses it by nameThe value is blanked out in what the bot reads and writes. |
| Internet only on the sites you allow? | YesOne list per agent. Every refused site is noted in the log. | With the proxy onA list in the proxy, which is off by default. | One list for everyoneA single list for the whole gateway. Without it, everything is allowed. | Enterprise onlyWithout a policy, everything is allowed. |
| Can you set a spending limit on each agent? | YesOnce it hits the limit, it stops and lets you know. On the business plan, per team too. | Not in moneyLimits turns and time per conversation. In money, only for paid tools. | Not in moneyA token cap per goal. Model spending only shows up on the dashboard. | Not yetThe whole account's limit applies. |
| Does it run on your own server? | YesAll of it. Open core, under Apache 2.0. | YesAll of it. Open source, under MIT. Here we're the same. | YesAll of it. Open source, under MIT. Here we're the same. | The page doesn't say |
The company wants a crew of agents that works for the whole team, with nobody writing code. Each agent in its own corner, with a spending limit and a record of everything it did, on your own server.
A technical person wants their own agent, one that learns from them and runs on any machine.
Each person wants their own assistant on WhatsApp or Telegram and is happy to look after it alone.
Each person wants their own assistant, and the team wants to talk to it on Slack.
A good agent is one you can leave to work in peace. That's why each one has its own space, only sees what's its own and only goes out to the internet if you let it. Company keys are stored encrypted. Whoever comes in, comes in by invitation. And everything is noted down, for anyone on the team to check whenever they want.
An AI colleague with a name, a role and a space of its own to work in. Behind it runs Claude Code or Codex.
No. Creating an agent means giving it a name, picking the team and saying what it's for. The conversation happens in the chat.
No. Each one works in its own space. Only those you put in the same space see the same material.
Only if you let them: by being in the same group or with permission. Every attempt is recorded. Direct conversation between agents in the chat is coming soon.
Admins look after everything, including people, the vault and limits. Members chat and create agents. Viewers follow everything without changing anything.
With an invitation link that lasts 24 hours and can only be used once. No password to remember. On the business plan you can also sign in with Google Workspace or another company login, and only people an admin invited get in.
On your own server. Trupe runs on your infrastructure.
Claude Code and Codex. Each agent can use the subscription a person already has or the company key, kept in the vault.
Every agent that uses the company key has its own spending limit, and on the business plan the whole team can have one too. When it gets there, the agent rests, nothing is lost, and an admin decides whether to raise it.
Yes. Each agent picks the conversation up where it left off and has a memory of its own that grows with use. The team can read everything it kept and correct it when needed.
The core is open, under Apache 2.0. Company login, per-team limits and the browsable log are part of the business plan, under their own license.
Start with one agent for one team. The others arrive when the work calls for them. Leave your details and we'll talk about your team.