Web
Assign work, follow progress, confirm results, manage the organisation. The left side shows your expert team and the list of items; pick someone and chat like it's IM.
- Nothing to install, and progress is readable on a phone
Niumate
Niumate turns AI into named experts with defined roles. You tell them what needs doing, they get it done on real work machines and real projects, and then bring the result back for you to confirm.
Put together this week's project progress report, only for project A, don't change any files, output Markdown and list what needs confirmation.
Goal · scope · limits · expected outputGot it. Filed under “Weekly progress report”, split into 3 work items. On it.
Report is ready. Two things need your confirmation: budget basis, next week's priorities.
Awaiting confirmationThree ways in
You mostly assign work from the web and Feishu; the desktop client is for people who contribute a work machine.
Assign work, follow progress, confirm results, manage the organisation. The left side shows your expert team and the list of items; pick someone and chat like it's IM.
DM the organisation assistant and just say what you need; or add it to a group, where it only responds when you @ it — other chatter won't disturb it.
Connect your own computer to the organisation so experts can work on it. macOS, Windows and Linux; Linux servers install as a background service with one line.
A request made in Feishu shows up in the web conversation and item list too — both sides see the same progress.
A typical run
You only take part in the first and last step — the middle does not need watching.
Pick an expert on the web home page and start a conversation, or tell the organisation assistant in Feishu. Not sure who to ask? The team lead takes it.
Once understood, it is filed into the right item and anything hands-on is split into work items. The timeline shows how many work items were created and who is on each.
The expert works on the machine in the background — reading project files, running tools. Progress is visible live and you can open the details.
If something is missing, the conversation prompts you. When the result looks right, confirm and the item moves on. Wrong direction? Stop it and restate what you want.
These are all the words the interface uses: what you ask for is a request, it gets filed into an item (a group of work around one goal), and the item is split into work items, each executed by one expert.
Where capabilities come from
Start from an existing role template, then swap the model, the machine or the capabilities as needed. Two employees on the same role with different models is fine.
Keys to external systems: databases, third-party API keys, granted per person. Built-in types not enough? Define your own.
Reusable how-to guides. Write your team's habits — release process, report format — down as a skill, tick it for the right people, and they all follow it.
Platform-provided connections (search, Feishu and more) plus your own MCP connections. Granted per person — a tool that was not ticked simply cannot be called.
Two come with every organisation
Fixed division of labour, neither crossing into the other. Display name, avatar, style, model and machine are all editable; the remit is not.
Rather not click through settings one field at a time? Message platform upkeep from the home page. It reads the current config, explains the blast radius, and checks with you before anything sizeable — high-risk operations always come back to you first.
Boundaries
Three hard rules are written into the built-in employees' remit. They are not switches you can turn off.
Anything involving payment, signing or authorisation is handed back. It is never decided on your behalf.
Reading files and running tools happen on the work machines you approved. Code repositories stay where they are.
Sensitive credentials go only to the employees that need them, and once stored the plaintext is never readable in the interface again.
FAQ
Finer operational questions all live in the user guide, available once you sign in.
Bring your own model account (BYOK) and pay the model provider directly — Niumate does not resell model usage. Common providers are supported, as are models on your internal network.
On your own work machines. Reading project files and running tools happen on the machine running the client, and code repositories stay where they are. For a stricter boundary, talk to us about private deployment.
Yes. Assigning work means saying a sentence on the web or in Feishu — no configuration involved. Model accounts and work machines are set up once by an admin, after which members only ever see a conversation.
Nothing is feature-limited during the 14-day trial: 10 members, 10 machines and the complete assign-and-confirm flow, with no payment info required. Model accounts are still your own.
Niumate