How we price AI in Taskeni

Taskeni has optional AI helpers: summaries of long threads, suggested steps for a task, meeting notes turned into tasks, a weekly recap, and soon an assistant and search by meaning. Each of those requests is a call to a model provider, and each call has a price. This post explains how we turn that price into what you see on your plan, so there are no surprises in either direction.

The short version: AI use is measured in credits, credits follow what each request actually costs, every plan has a monthly allowance, and AI stays off until someone in your workspace turns it on.

Where the cost comes from

Taskeni's AI helpers use Google Gemini. Gemini, like every hosted model, is billed by the token, roughly three-quarters of a word. Input tokens are what we send: the task, the comments, the instructions. Output tokens are what comes back, including the model's internal reasoning, which Google bills as output.

We use two Gemini models plus one for search:

• Gemini Flash-Lite for short, structured jobs: catching up on a task, the Home briefing, and turning notes into tasks.
• Gemini Flash for jobs that need more care: suggesting steps for a task and writing the weekly recap.
• Gemini embeddings for smart search, which matches tasks by meaning.

Google's list prices in October 2026, per million tokens, were $0.30 input and $2.50 output for Flash-Lite, $1.50 input and $7.50 output for Flash, and $0.20 for embeddings. Flash's list price is scheduled to rise on January 1, 2027; we already use the 2027 price in our calculations, so our pricing doesn't depend on a promotion ending. Check Google's Gemini API pricing page for current figures.

A single request is cheap. A whole team's requests over a month are not nothing, and some requests are 20 times the size of others. A flat "AI included" price would have to be set for the heaviest users, which means everyone else pays for them, or capped somewhere hidden. We chose to count.

What a credit is

One AI credit is $0.0024 of model cost at list price. That's about 1,000 tokens of typical Flash use.

After every request, Taskeni records which model ran, how many input and output tokens it used, and works out the credits from that model's price. Nothing is estimated in advance, and nothing is rounded up to a whole credit. The smallest charge for a request is a hundredth of a credit.

Some examples of what that means in practice, using typical sizes:

• Turning a page of meeting notes into tasks on Flash-Lite: about 1,500 tokens in and 400 out comes to roughly 0.6 credits.
• Catching up on a task with a few dozen comments on Flash-Lite: about 3,000 in and 300 out is roughly 0.7 credits.
• Suggesting steps for a task on Flash: about 1,000 in and 400 out, including reasoning, is roughly 2 credits.
• A weekly recap on Flash: about 2,000 in and 300 out is a little over 2 credits.

A long thread, a busy week, or (when it arrives) an assistant request that reads many tasks will use more. That's the point of measuring: a quick helper uses a fraction of a credit, and a big request uses several.

What each plan includes

Every plan has a monthly allowance that renews on the 1st.

• Free: 100 credits per workspace per month. That's a trial: enough to see what the helpers do, not enough to run a team on.
• Pro: 300 credits per member per month, pooled across the workspace. A ten-person team has 3,000 credits to share, so one heavy user doesn't run out while others barely use theirs.
• Business: 1,000 credits per member per month, pooled.

In model cost, a Pro member's 300 credits is $0.72 a month, and a Business member's 1,000 credits is $2.40. Those figures are built into the plan prices, along with hosting, email, storage, payment fees, and a share of the people using the Free plan. Pro and Business are on sale now, and Pro starts with a 14-day free trial. See pricing.

When a workspace uses its allowance, the AI helpers pause until the 1st, and everyone in the workspace sees a short notice saying so. Nothing else in Taskeni changes: boards, search, integrations, and the MCP server keep working. Owners and admins can see the exact count (credits used this month, the allowance, and any top-up balance) in Settings → Workspace.

Top-ups

On any plan, owners and admins can buy top-ups in Settings → Plan & billing: 1,000 credits for $4, 5,000 for $20, or 10,000 for $40. Top-up credits are only used after the monthly allowance runs out, and they don't expire.

Here's the arithmetic. 1,000 credits is $2.40 of model cost. Our payment provider takes a fee of roughly 7% plus a fixed amount per purchase. What's left after that is a little over 30% above the model cost. That margin is fixed: it's the same rule we use for plan prices, where every price has to stay at least 30% above what it costs us to serve after payment fees. We check this with an automated test, which fails if a price or the top-up ever drops below that line. If Google's prices change a lot, we'll change the credit size or the top-up price and say so on the changelog.

AI is off until you turn it on

New workspaces start with AI helpers off. A workspace owner or admin turns them on in Settings → Workspace. Until then:

• No workspace data is sent to Google.
• Search stays keyword-only.
• The AI buttons don't appear in tasks or on Home.

Turning AI off again stops all requests immediately.

What we send, and what we don't

Each request sends only what it needs.

• Catch me up on a task: the task's title and description (up to about 4,000 characters), and up to the last 60 comments and 60 activity entries, with the names of the people who wrote them.
• Catch me up on Home: up to 250 recent activity lines from boards you can see, with task titles and board names.
• Suggest steps: the task's title, description, board name, and existing checklist items.
• Notes into tasks: the text you pasted (up to 12,000 characters), the names of people who can see the board, its labels, and today's date.
• Weekly recap: the week's counts, board names, and the titles of up to five finished and five slipped tasks per board.
• Smart search: the text of tasks and comments, to build the search index, and the words you search for.

We never send passwords, the contents of files, or anything from boards you can't see. Requests run as you, so the AI can't read anything you couldn't open yourself.

Content from your workspace is wrapped and labeled as data, and the model is told not to follow instructions that appear inside it. That reduces the chance that a pasted email or a comment can steer the output. Results are also checked before you see them. Summaries must cite real comments or tasks, or the point is dropped; suggested assignees must exactly match a real person on the board; dates must be valid. And nothing the AI suggests is applied without you: you review the steps, the tasks, or (with the assistant) the list of changes, and click to apply.

Using your own AI instead

If you already pay for Claude, ChatGPT, or Cursor, you can connect it to Taskeni through our MCP server. The assistant then runs on your own account, and Taskeni only answers its tool calls, so it doesn't use Taskeni credits at all. The MCP server is on every plan.

What we won't do

We won't advertise "unlimited AI", because every request costs something and "unlimited" would mean a limit you can't see. We won't turn AI on for you. And we won't change how credits are counted without saying so first.

If something here is unclear, or you think the numbers are wrong, write to support@taskeni.app. More detail on the helpers themselves is on the AI feature page, and plan details are on pricing.