Skip to content
Tally

The ops agent for your AI and SaaS tools

Your team's new toolbox, finally under control.

Tally finds every paid AI tool, tracks seats and tokens in one place, and shows what the spend actually produced. It measures tools, not people.

Join early access

How it starts

It starts with a few reasonable yeses.

Four people, four fair requests, four ways to be billed.

  1. I'll let the agent finish tonight.Backend EngineerTokens, Claude Code
  2. I put ChatGPT on my card, just for this project.Product ManagerSeat, personal card
  3. Marketing bought their own Figma seats.Design LeadSeat, duplicate
  4. We ran out of credits on the 18th, so I topped up.Finance LeadCredits, top-up

The sum

Then it all lands on one card.

Nobody meant to spend it. Every line was a yes.

78%

of IT leaders were hit by unexpected AI or usage charges.

Zylo 2026 SaaS Management Index

Card statementAI tools, all teams

DateMerchantAmountOwner
10/01OPENAI *CHATGPT BUSINESS$1,200Unknown
10/01CURSOR AI SAN FRANCISCO$600Unknown
10/03CURSOR AI SAN FRANCISCO$360Unknown
10/04ANTHROPIC *CLAUDE TEAM$750Unknown
10/06GITHUB *COPILOT$570Unknown
10/12OPENROUTER INC NEW YORK$1,460Unknown
10/14OPENAI *CHATGPT PLUS ×11$220Unknown
10/20OPENAI *API USAGE$1,180Unknown
10/22NOTION LABS INC$800Unknown
10/28PERPLEXITY AI$400Unknown

Total$7,540

The turn

Now make it tally.

The weekend loop, caught before the bill.

Every project's month, live. One of them isn't like the others.

Team sessionsLIVE

Showing the last 30 days.

$3,340448 sessions · equivalent API value$300 through the gateway · $1,160 refused at the cap

ProjectAPI valuevs usual
mobile-app$1,120+4%
docs-site$740−2%
checkout-web$610+8%
internal-tools$390−5%
data-pipeline$310+3%
support-agent$170Anomaly: 10× usual

support-agent hit 80% of its $300 budget.Tally toSlackSlack · Sat 11:19 AM

What Tally does

It doesn't wait for you to look. It tells you.

Tally is not another dashboard. It is an ops agent: it checks your tools and bills continuously and comes to you, in Slack or email, with specific findings and a next step.

  1. Seats and bills

    ChatGPTCursorClaude CodeGitHubNotionPerplexity

    Tokens and sessions

    Claude CodeCodexOpenRouter

    What shipped

    VercelGitHubFigmaLovable

    + 9 more tools

  2. Tally checks, all month

    • Idle seats
    • Duplicate plans
    • Runaway loops
    • Budgets at 80%
    • Hard caps at the gateway
    • Spend next to what shipped
    SlackCodexClaude CodeFigmaVercelNotion

  3. It comes to you, in Slack or email.

    TallyAPPSat, Oct 10, 11:19 AM
    ⚠️ Project support-agent · Gateway: reached 80% of budget ($240 of $300)

    Project support-agent · GatewayNear limit
    Refused since Sat, Oct 10, 1:35 PM$1,160
    Seats and every other tool: working

The whole toolbox

One answer to “what are we paying for, and what did it produce?”

Beyond tracking, alerts and caps, Tally takes in the rest of the toolbox: the seats, the credits and what they made.

  1. Step 01

    ChatGPT Business
    40 seats · $1,200
    renews Nov 1
    told 7 days early
    11 Plus plans on personal cards
    $220

    Every paid tool, explained before renewal

    Every seat, on the company card or a personal one, before it renews.

  2. Step 02

    Cursor Teams
    24 seats · $960
    9 idle for 30 days
    $360 / month
    Perplexity Ent. · 10 seats, nobody uses
    $400

    Cut the waste

    Find idle seats and duplicate tools across teams before your next renewal.

  3. Step 03

    Claude Code + Codex
    448 sessions · $3,340
    mobile-app · docs-site
    2 projects
    checkout-web · internal-tools
    2 projects

    Spend next to what it made

    Each tool's cost next to the sites, apps, docs and designs your team produced, not just the code.

Integrations

Not just the coding tools.

Seats, credits and what they shipped for the whole toolbox, next to Claude Code, Codex and the model APIs.

LovableLovablePlan and monthly credits; the apps it published

Privacy boundary

We measure tools and output, not people.

Reports are team-level by default. No screenshots, no keystrokes, no rankings. The gateway never stores what passes through it.

Never sent to Tally

  • Not sent: Prompts and responses
  • Not sent: Tool calls and their output
  • Not sent: File contents and paths
  • Not sent: Source code

TallySent to Tally

  • Sent: Token counts per model
  • Sent: Timestamps and message count
  • Sent: Tool and session id
  • Sent: Project label and branch

FAQ

Straight answers.

Who is Tally for?
Teams of 20–150 that took up AI tools fast, on company and personal cards, without an IT department tracking them, and the founder or COO who has to answer what it all costs and what it produced.
Does Tally monitor my employees?
No. We measure tools, spend and output, not people. Prompts, responses, code and files stay on the computer. Reports are team-level by default; nobody is ranked.
Is this another dashboard?
No. Tally is an ops agent. It checks your tools and bills continuously and comes to you, in Slack or email, with specific findings and a next step: budget alerts, idle seats, duplicate tools and spend with no output behind it.
How is this different from Anthropic's, OpenAI's or GitHub's own usage reports?
Each of them reports its own tool, and more of them now report per person. Tally puts every tool on one view for the whole team, seats and tokens together: Claude Code, Codex and OpenRouter next to Figma, Notion, ChatGPT and the rest, with budgets and alerts across them and a hard cap on gateway traffic. It never reads prompts.
Which tools does it work with?
Claude Code and Codex (the CLI and the Tally Mac app), OpenRouter and any tool routed through the Tally gateway, with alerts in Slack and email. The Mac app also finds AI tools installed on each computer. And the rest of the toolbox: Lovable and Vercel, Figma, Canva, Notion, ChatGPT Business, HubSpot, monday.com, Linear, Slack AI, Midjourney, Adobe Firefly, Framer and Webflow, plus Claude and OpenAI admin-panel spend.
Is there a Mac app?
Yes. Tally for Mac (v0.1.1, Apple Silicon and Intel) lives in the menu bar, pairs your computer, finds your AI tools and syncs usage in the background. For now it's for invited teams: ask your team admin for the installer. Prefer the terminal? Developers can also connect Claude Code and Codex using the Tally CLI (tallyless).
How do I get access?
Tally is in private beta. Join the early access list and we'll email you when there's a spot.
Why are seats and tokens counted separately?
Many tools are both: a plan per person plus a credit pool that runs out.
How does the seat check work?
Billed: AI subscription invoices from the admin's email, or uploaded by hand. Reported: read-only admin keys from OpenRouter, Claude and OpenAI. Used: what each person actually uses, measured on their own computer. The three are checked against each other.
What's the difference between equivalent API value and actual spend?
Equivalent API value: what subscription use would cost at list prices. Actual spend: money really paid. The two are never added together.
Where does our data live, and who can see it?
Each team is its own tenant, isolated from every other team, and provider keys are encrypted and bound to their team. The details are in the Privacy Policy.
Is our data kept apart from other teams'?
Yes. Every team is its own tenant: teams are isolated from each other, every query is scoped to the caller's team, and the isolation is covered by automated tests.
How are our provider keys stored?
Encrypted, and each one bound to its own team. The same goes for webhooks. Admin keys are read-only.
Does Tally rank people by usage?
No. Insights and streaks are team-level. Nobody is ranked by tokens or spend; people see their own usage and the team's totals.

Tally is in private beta.

We're letting teams in a few at a time. Request access and we'll email you when there's a spot.

We'll use your email only to contact you about access, as our Privacy Policy describes.