Tasklet vs Kortix: the open‑source option you keep

The first hour: what Tasklet hands you before lunch

Tasklet’s homepage makes a large promise: "Automate any business task with AI agents that run 24x7." The setup is built to keep it. You sign up, describe a job, connect the tools, and the agent runs in Tasklet’s cloud while you do something else. No install, no model key to fetch, no server to patch on a Sunday.

The first hour is useful, and easy to forget after a year of running infrastructure. An ops lead can have an agent reading a shared inbox and drafting replies before the morning standup ends. Pricing is public on tasklet.ai/pricing. A free tier carries limited trial credits and no card. Then plans from $25 a month for casual use, $100 for power use and a $250 tier, with custom amounts up to $10,000. Starter at $25 carries 10,000 credits, Pro at $100 carries 40,000, and the $250 tier carries 100,000. Billing sits at the organization level, with unlimited team members on one shared credit pool, so headcount does not move the bill.

That speed rests on one decision, and it is the one you make with a click. Everything that shapes the agent stays on Tasklet’s side: the prompt, the tool list, the permission rules. So do the vault holding the connection keys and the record of what each run did. For a team whose work is contained, that is a fair price for a working agent today.

The bill for that decision arrives later, in a week nobody scheduled. Someone asks why the agent that worked in March behaves differently in June. You have the run history and no diff. Someone else asks what the agent was allowed to do when it emailed a customer, and the answer is a settings screen that has changed since. The run history shows what happened. The reasoning behind each permission change lives in a screen nobody can diff.

The third year: a repo, a change request, and keys that stay put

On Kortix an agent is a block in a file in a git repo you own. The starter template shows the shape in six lines:

default_agent: worker
agents:
  worker:
    mode: primary
    permission: allow
    prompt_file: agents/worker.md
    tools: all

The default agent there is worker. Its permission is allow, the loose setting, and it is the first line a careful team tightens. Its prompt sits in a markdown file at agents/worker.md, so the words that shape its judgement have a diff and a history. Naming its tools one at a time is an edit to a single line. None of it needs a dashboard, and none of it disappears when a plan is repriced.

The interesting part is how a change lands. Session work reaches main through a change request. Merge is default-deny for agents. An agent can propose a tighter permission, a new skill or a rewritten prompt, and a person reads the diff before it merges. Approval gates you set. Off until you set them. The permissive setting is where every project starts, so the first week is when a team decides which actions need a person.

Credentials get asked about last and regretted first. Kortix’s docs put it plainly: "Credentials stay on the Kortix server and never enter the agent’s computer." An agent can call the invoicing system without carrying a key to it. On Tasklet those live in one product, held and patched by the vendor. That is also why the approval flow works without your team running anything.

Where it runs is a setting rather than a tier. Any model, your own keys, on Kortix Cloud, in your VPC or on your own on-prem network. The install is a script. A self-hosted instance reads the same config a managed project uses, so the second machine is a copy rather than a migration. Open source. Read it, fork it, audit it. The code is on Kortix on GitHub.

Who should pick which

If nobody on your team wants to be on call for a server, Tasklet at $25 a month is cheaper than the first afternoon of an install. The agent also works today. Choose Tasklet when the work is contained, the tools are common, and the agents never touch money, customer records or code. That is a large share of small teams, and the product serves them well.

Choose Kortix when the agents will reach the invoicing system, the customer database or the codebase, and when somebody will eventually ask who allowed a change. The repo answers that with a diff, and the person who merged the change request has a name attached. The same files keep a security review short. A reviewer can read the code before it gets a credential, and the permission block before it gets a key.

Pick for the questions that arrive in year three, and for whether the machine that answers them is one you keep.