Someone asks in Slack, Linear or GitHub. Agentick works out the job, briefs Claude Code or Codex on a Mac you own, checks the result, and replies where you asked. Every step on the record.
Already have a workspace? Sign in
@finance-agentCan you reconcile October's vendor invoices against the ledger before Friday's close?
Triaged as work. September's reconciliation found and attached for reference.
Invoice inbox export, the ledger, last month's workbook and the close checklist.
Matching 212 invoices to ledger entries, flagging anything that doesn't tie.
Workbook totals match the ledger. Every flag has its source invoice attached.
Reconciliation · Oct 2026.xlsx in the Finance drive, flagged items listed first.
Done. 209 of 212 invoices match. Three are flagged with evidence in the report; Acme's looks like a duplicate.
How it works
We run the loop around the agent: picking up the work, briefing it, checking it, reporting back where you asked. The agent itself is the one your team already uses, and the work can be anything a person at a computer would do.
Two surfaces
Desktop runner · macOS
Enroll it with a token from your workspace, pick the engine, and leave it running. It does nothing until a case is briefed to it.
brew install --cask metapouch/tap/agentick-desktopDownload for MacWeb dashboard
Cases show their state plainly: working, waiting on a person, waiting for a computer, done. No digging through logs to find out where things stand.
Engines
Claude Code and Codex are general-purpose agents that happen to be very good at code. They read files, run scripts, use a browser and draft documents. Agentick doesn't resell their usage; prompts go from your Mac to your provider, on your plan.
On your own Claude subscription. Sessions stream into the runner with tool calls, diffs and permission prompts you can answer from anywhere.
On your own Codex login. Same brief, same checks, same hand-back at the end. Switch engines per agent whenever you like.
Permissions and audit
Agents do real work in your systems. These are the answers a security review asks for, published rather than promised.
Agents run on a machine you own, using the logins already on it. We hold no credential to your repositories.
Prompts and completions go from your machine to your model provider directly. They don't pass through us.
Each case keeps a timeline of what the agent did, who approved what, and the evidence behind every decision.
Usage and cost
3 agents worked this month. The idle one used nothing.
Pricing
You bring the Mac and the Claude Code or Codex subscription. We don't mark up model usage.
$0
$19 / month
or ₹999 / month + GST
$29 per active requester / month, minimum 3
$49 per active requester / month, minimum $500 / month
Custom, from $25k / year
An active requester is a person who opened at least one case that month. People who asked nothing that month cost nothing.
A turn is one message that sends an agent back to work. Prices in rupees are shown before GST, which is added at checkout. What each plan includes.
Waitlist
We're onboarding founding teams a few at a time. Tell us where to reach you.
FAQ
No. The first agents were built for engineering, but a case can be anything a person does at a computer: reconciling invoices, screening candidates, answering support tickets, preparing a board pack. The agent gets a brief, a Mac and the logins it needs.
No. The runner works on any Apple silicon Mac running macOS 13 (Ventura) or later. Many teams start with the Mac mini already sitting under a desk.
Claude Code and Codex, each on your own subscription or login. You pick the engine per agent and can switch at any time.
Case metadata, the brief, check results and the timeline. Prompts and completions go directly from your Mac to your model provider. The full list is in What leaves your network.
Every agent is granted a specific set of tools, and anything outside that grant is refused by default. You can set an agent to ask instead: the request goes to a person on the case, who allows or denies it from the case in Slack or the dashboard. If nobody answers within ten minutes it is denied, so an agent never waits on nobody.
A person who opened at least one case in a calendar month. Team and Business are priced per active requester, so people who asked nothing that month cost nothing.
Solo is bought in the dashboard, under Settings → Plan, in rupees by card or UPI with GST added. Payments are processed by Razorpay and renew monthly. Cancel any time from the same page and the plan runs to the end of the period you have paid for. Team, Business and Enterprise are arranged with us and invoiced.
Yes. There's a runner CLI: curl -fsSL https://console.agentick.xyz/install.sh | sh, or brew install metapouch/tap/agentick-runner. The setup guide covers both.