New: AI spend limits now cover the agent builder and the project planner →
Product
AI Teammates
Solutions
Resources
PricingEnterprise
Start free Book a demo Log in
AI teammates

AI teammates that do real work, and ask before they act.

Assign them tasks, @mention them in comments, put them on a schedule. Each one works in a sealed sandbox, shows every step, and waits for your OK before touching your other tools.

2 AI teammates on the Free planAI teammates are never seats
app.taskcrew.com
A task with a finished run of the Blog & Article Writer: inputs read from the task, tool calls, tokens and cost
A real run: inputs from the task, each step, tokens and cost
What a teammate is

A role, the inputs it needs, and the tools it may use.

Start from one of the templates, or describe what you need to the agent builder and it designs one for you.

A role

What it does and how it works, written as instructions: “Draft a weekly status update from what happened in the project.”

The inputs it needs

Declared fields it reads from the task, like topic, audience or period. A missing required input stops the run and says exactly what to add.

The tools it may use

Workspace tools (tasks, comments, wiki pages, search) plus the connectors you allow, each read-only or asking first.

Or just describe it.

“Every week, summarize our Linear issues into a status report.” The agent builder asks a few questions, checks which tools it can use, and builds the teammate. You can fine-tune everything later.

  • Definitions are versioned
  • Every run keeps an exact snapshot of the definition it used
  • Hand it to your team in one click
app.taskcrew.com
The agent builder chat with suggestions such as “Research a company before a sales call”
The agent builder, in the app
Seven ways to start one

Delegate the way that fits the work.

Assign + Start now
@mention in a comment
A button on the task
A schedule
When a task changes state
An automation
A recurring task

An @mention in a comment is the instruction. A teammate @mentioned in a thread answers in that thread.

Watch the run

Every step, live, on the task.

What it read, which tools it called, how long each took and how many tokens it used. Cost is shown to owners and admins only.

Agent Working→ Agent Done · needs review or Agent Failed · reason + Retry

Try approving or rejecting

🕵️
Lead Researcher DEALS-14 · Acme Corp — intro call
Agent Working
Sandboxed run·0 tokens·0.0 s
  1. Lead Researcher started working
    Assigned by Sara · Start now
  2. Checking connectors and preparing tools
    Exa Web Search · HubSpot
  3. Lead Researcher wants to create note in HubSpot
    Waiting for your approval
A replay of a real run's steps. Try both buttons.
Teammates that work together

A lead teammate can hand out the parts.

It splits the work into subtasks and assigns each to the right teammate. Results flow back to the parent task. How deep and how wide it can go is limited, so nothing runs away.

🎯
MKT-40 · November launch campaignCampaign Brief Planner plans it and hands out the parts
📣
Launch posts for LinkedIn, X and InstagramSocial Media Post Pack
🧭
Keyword brief for the launch articleSEO Content Brief
✍️
Draft the launch articleBlog & Article Writer
Each result comes back to the parent task for a person to review
You stay in control

Reading is free. Acting waits for a person.

Asks first

Each action in Slack, HubSpot, Gmail and other connected tools waits for Approve or Reject, with the exact details.

Sandboxed

Every run gets its own sealed container: no network, no secrets, a read-only file system.

Spend limit

Set a monthly limit. At the limit AI pauses until you raise it or the month ends. No surprise bill.

Every step logged

Every tool call, with its inputs, results and timing, is written to the audit log by the runner, not the AI.

Sees what you share

Ask your workspace answers only from what the person asking can open.

Never sees passwords

Password vaults are invisible to AI teammates, MCP, search and exports.

Read how approvals work

Built-in AI features

AI where you already work.

Ask your workspace

Ask “what’s overdue for Acme?” and get the tasks that answer it, using only what you can open. It never changes anything. 20 questions per person per hour.

Describe your project

A few sentences in, a planned project out: lists, workflow, fields, starter tasks, recurring work, a wiki, a playbook and suggested teammates. Untick what you don’t need.

Status reports

The Project Status Writer drafts Headline · Done · In progress · Risks · Next week on a schedule. A person reviews it; nothing is sent to a client.

The Assistant

Ctrl/Cmd + J opens an Assistant beside any page. It knows where you are and helps from there.

29 connectors

Teammates can work in your other tools.

Connect once with OAuth. Tokens stay encrypted on the server, and teammates never see them. Add any remote MCP server as a custom connector.

NNotionJCJira & ConfluenceAAirtableGDGoogle DriveLLinearMmonday.comAAsanaHHubSpotIIntercomSSlackGGmailGCGoogle CalendarGCGoogle ChatCCanvaWWebflowWWixBBufferMAMeta AdsSStripePPayPalSSquareXXeroZZapierGGitHubSSentryCContext7DDeepWikiMLMicrosoft LearnEWExa Web Search

See all integrations

Credits included in every plan

From 500 a month on Free to 1,800 per person on Business, pooled for the workspace.

AI pauses at your limit

Never surprise overage. Runs under way finish; new ones wait for the 1st.

Spend you can see

By teammate, project, person and month, with alerts at 80% and 100%.

How AI pricing works

FAQ

Questions, answered

Which model do teammates use?

Every AI teammate, the agent builder and Describe your project use deepseek/deepseek-v4-flash through OpenRouter. By default, requests only go to model providers that don’t keep or train on prompts.

Can a teammate send emails or post on my behalf?

Only through a connector you have connected, and with “Ask before acting” on (the default for new teammates), every action that changes something outside TaskCrew waits for a person to approve it, with the exact arguments shown. Reading is never blocked.

What can’t teammates see?

Password vaults, other people’s My space, and anything the person who started the run can’t open. Their sandbox has no network and no secrets: the keys stay with the runner.

What happens if a run fails?

The task moves to Agent Failed with the reason in plain words (for example, a connector that needs signing in again) and what to fix. Retry starts it again. A run is never repeated silently, because a tool call may already have done something.

Do AI teammates count as seats?

No. AI teammates are free. Plans differ only in how many you can have: 2 on Free, 10 on Team, unlimited on Business and Enterprise.

Can I write my own teammate?

Yes. Describe it in the agent builder chat, or start from a template and change its role, inputs, tools and connectors. Definitions are versioned, and every run keeps an exact snapshot of the version it used.

Give your first task to an AI teammate today.

Free for up to 5 people, with 500 AI credits a month and 2 AI teammates.

No credit cardGuests and AI teammates are freeOr self-host with Docker