Create an AI agent from one sentence.
To create an AI agent in Arclyx, describe the job in one sentence. Arclyx asks a couple of quick questions, then drafts the agent: its instructions, the tools it may use and, if you name one, its schedule. Try it once on your real tools, change it in plain words, and let it run.
Inbox triage
- Read today's unread Gmail, leaving out newsletters and receipts. Sort each one into Needs you today, Can wait or FYI, with one line on why. Send the list as one email.
- Gmail · readEmail · send to me
- Weekdays · 8:00
- Arclyx picks a model for each run
How do you create an AI agent?
Five steps. Only the first asks you to write anything; the rest is reading, trying and deciding.
- 1
Write the job in one sentence
Say what it should do, with what, and when. Plain words; no code, no flowchart.
- 2
Answer a couple of questions
Arclyx asks at most three things it can't work out, each with an answer already picked.
- 3
Read the draft
Instructions, tools and schedule, written for you. Change anything by saying what should change.
- 4
Try it once on your real tools
A dry run: it reads with your real tools, and anything it would post or send is shown to you instead. Every step shows as it happens.
Test runs → - 5
Let it run
On a schedule in your time zone, from a webhook, when someone @mentions it in Slack, or whenever you ask.
Triggers →
What does creating an agent look like in Arclyx?
This is the New agent screen: the sentence goes in the box, or you start from a ready-made job. Arclyx drafts the agent on the right as it reads, and nothing runs on its own until you create it.
AgentsNew agent
New agent
+ Set it up yourselfWhat should it do?
Every weekday at 8, read my unread Gmail, sort it into what needs me today and send me the list.
Or start from a job
Inbox triageSorts the day's unread mail into what needs you
Meeting briefA brief before every external meeting- Competitor watchWhat competitors changed this week
Prospect finderProspects with a reason to reach out now
Drive librarianFinds the file you half-remember
PR scoutSummarises open pull requests
Reading the job
Inbox triage
Sorts today's unread email into what needs you.
Tools
Gmail
Runs
Weekdays at 8:00
Drafting saves a draft. Nothing runs on its own until you create it.
How do you write a good sentence for an AI agent?
Name four things: the job, the tools it works in, when it runs and where the result goes. Arclyx asks about anything you leave out, but a sentence that names all four gets a draft that's right the first time. Be specific about the job; leave the how to Arclyx.
- What done looks like, in plain words: “sort into what needs me today”, not “help with email”.
- Where it reads and writes: Gmail, HubSpot, a Notion page, the web.
- Every weekday at 8, the first Monday of the month, or leave it out to run on demand.
- An email to you, or a Slack channel.
“Help me with my email.”“Every weekday at 8, sort my unread Gmail into what needs me today and email me the list.”
Says what done looks like, where to look, when, and where to send it.
“Keep an eye on competitors.”“Every Monday, check what Competitor A and Competitor B changed on their websites last week and post a summary in #marketing.”
Names who to watch, the window, and the channel.
What goes into an AI agent?
Every agent has the same parts. One sentence names most of them; Arclyx fills in the rest and shows you each one.
Every weekday at 8, read my unread Gmail, sort it into what needs me today and send me the list.
- A schedule, in your time zone. Leave it out and the agent runs when you ask.
- Only the tools that job needs, through your own sign-in. Here, Gmail to read it.
- Instructions the agent follows on every run, written for you and yours to edit.
- Where the result lands: an email to you, or the one Slack channel you choose.
- Spend limits checked before every run: your plan's, plus a ceiling per run and a monthly cap on paid plans.
What does every agent come with?
Each part has its own page with the detail.
Try it before it goes live
Run a draft once on your real tools and watch every step and tool call as it happens.
Test runs →Runs on its own
Daily, weekdays, weekly or monthly, from a secure webhook, when someone @mentions it in Slack, or on demand.
Triggers →Works in your tools
Gmail, Drive, Calendar, Outlook, Notion, GitHub, HubSpot and more, each with your own sign-in. Admins can add any remote MCP server.
Connectors →Stays inside the lines
Only the tools you grant, spend limits on every run, email only to you, code in an isolated sandbox.
Guardrails →Shows its work
Every run recorded step by step, with tool calls, tokens and cost, and a history chart per agent.
Observability →Ask about any agent, from its page
Ask about the agent, run or project in front of you, say “show me” to open an agent or run, or describe a job to draft an agent.
Ask Arclyx →
What else can an agent work with?
Beyond your connected tools, an agent can read the documents you upload, make real files, and share a big job with helpers.
Answers from your documents
Upload PDF, Word, Excel, CSV or text files. Agents search them and cite the source, and access tags decide who can see which documents.
Real files, checked before they're sent
Agents write code in an isolated sandbox to make PDFs, spreadsheets, Word documents and charts, then look at what they made before handing it over.
Agents that share the work
On Team, an agent can hand parts of a job to helpers that run alongside it, then review what comes back before it answers.
Can you make an AI agent without coding?
Yes. In Arclyx the whole agent comes from what you write: the sentence, your answers, and any change you ask for in plain words. Nothing to install and no flowchart to wire. If you want control, every part is a setting you can open: the model, how many rounds it may take, its spend limits and its instructions.
- Arclyx picks one per run, or you choose
- How many steps a run may take
- A ceiling per run and a monthly cap (paid plans)
- Email to you, or one Slack channel
- Let it hand parts of the job to helpers (Team)
Start from an agent that already works.
The New agent page offers dozens of ready-made jobs. These run in Arclyx today; each page shows what it reads, what it posts and the sentence behind it.
What can an agent you create do, and not do?
It gets exactly the access you gave it and nothing more. Every run is recorded, so you can see what it did.
It can
- Use only the tools you granted, through your own sign-in
- Write email as drafts for you to send, and create docs and pages you asked for
- Email you, or post in the one Slack channel you chose
- Run code in an isolated sandbox with no internet access
It can't
- See anything your account can't
- Email anyone but the person it runs for
- Pick a new Slack channel on its own
- Spend past its limits
Questions about creating an AI agent.
How do you create an AI agent?
Describe the job in one sentence: what it should do, with which tools and when. Arclyx asks up to three quick questions, drafts the instructions, tools and schedule, and lets you try it once as a dry run on your real tools before it goes live.
Do I need to code to make an AI agent?
No. The agent is built from what you write, and you change it the same way, by saying what should change. Every setting is also there to edit by hand if you prefer.
What tools can an agent use?
Gmail, Google Drive and Calendar, Outlook Mail and Calendar, Notion, GitHub, Linear, HubSpot, Apollo, Clay and Cloudflare, each connected with your own account. Admins can add any remote MCP server. Agents can also search the web and your uploaded documents.
Which AI model does an agent use?
By default Arclyx picks a model for each run. You can choose one for an agent instead.
Can I start from a ready-made agent?
Yes. The New agent page offers dozens of ready-made jobs, such as inbox triage, meeting briefs and competitor watch, and names any tool you need to connect first.
Is it safe to give an agent my tools?
An agent only gets the tools you grant, through your own sign-in. It emails only you, posts only in the Slack channel you chose, stays inside its spend limits, and every run is recorded step by step.
Create your first agent.
One sentence in, a working agent out. Join the beta and we'll set you up.