One agent, for every task

aixplain One does your work and builds agents and apps in plain language, on any model.

Using Athena, which picks the right model for each step

Free for Mac, Windows and Linux

Hand it your work

It works in your apps and files, builds apps and writes code

Just ask

Ask in plain words and start from the outcome. It plans, connects what it needs and gets it done. It asks before it sends, pays or deletes, until you let it run on its own.

Desktop, the web, WhatsApp and Telegram

Works in your apps and files

It works with the files on your computer and the apps you connect: Gmail, Calendar, Drive and more, from 1,000+ tools, or your own MCP servers and APIs. Hundreds of skills are ready to use, and you can bring your own.

Browse the tools

Athena picks the right model for each step

Athena, aixplain's model router, sends hard steps to a frontier model and routine ones to faster, cheaper models. Or choose your own with Automode.

AutomodeAthenaCustom

DeepHard problemsClaude Opus

StandardRoutine workGLM 5.3

QuickConversationQwen · on-prem

Each answer shows the model that gave it

export async function POST(req: Request) {
  const { message, clientId } = await req.json()
  // The agent inside the booking app
  const run = await fetch(`${API}/v2/agents/booking-assistant/run`, {
    method: "POST",
    headers: { "x-api-key": process.env.AIXPLAIN_KEY },
    body: JSON.stringify({ query: message, session: clientId }),
  })
  return Response.json(await run.json())
}

describe("booking assistant", () => {
  it("moves a session to the next free slot", async () => {
    const res = await call("Move Thursday to Friday")
    expect(res.status).toBe("done")
    expect(calendar.events).toHaveLength(1)
  })
})
export async function POST(req: Request) {
  const { message, clientId } = await req.json()
  // The agent inside the booking app
  const run = await fetch(`${API}/v2/agents/booking-assistant/run`, {
    method: "POST",
    headers: { "x-api-key": process.env.AIXPLAIN_KEY },
    body: JSON.stringify({ query: message, session: clientId }),
  })
  return Response.json(await run.json())
}

describe("booking assistant", () => {
  it("moves a session to the next free slot", async () => {
    const res = await call("Move Thursday to Friday")
    expect(res.status).toBe("done")
    expect(calendar.events).toHaveLength(1)
  })
})

Start in aixplain Desktop, carry on in the CLI

The same agent runs on your computer, with your files and your models. It writes, tests and reviews code, and in your terminal it keeps the same permissions and sessions.

How coding works on Desktop

Works with your files

Open a folder and describe the task

Bring your own model

Your subscription, your API key or a local model

Stays on your computer

Sessions and settings are stored locally

Review every change

It asks first, then you read the diff and commit

Free to start, then pay as you grow

When you register, you get free credits for 7 days, no card needed.

See pricing

CoreFreefor free models, pay as you go for the rest

Scale$75per seat a month

EnterpriseCustom

The agent factory

The same agent that does your work builds new ones. It designs, builds, tests and deploys each agent, then monitors and improves it on autopilot, with evidence at every step.

Develop and deploy

Describe the work in plain language. aixplain One designs the solution with you and builds the agent, or you build it in code with the Agents SDK. It's tested on your machine, then deployed with its own API and key, versioned, on cloud or your servers.

aixplain One

Build an agent that books, moves and cancels sessions, and deploy it on our servers

Wrote agent.py with the calendar and payments tools
Ran 12 test conversations, all passed
Deployed v1 with a REST API and MCP endpoint

booking-assistant is live. Its key is in Settings.

Ask anything

booking-assistantagent.py

1from aixplain import Aixplain2aix = Aixplain()34agent = aix.Agent(5  name="Booking assistant",6  instructions="Book and move sessions.",7  tools=[calendar, payments],8).save()  # deployed with an API910agent.run("Move Thursday to Friday")

LivePOST /v2/agents/booking-assistant/runv1 · your servers

Monitor

Every run is traced: what the agent was asked, each step, how long it took and what it cost. Set thresholds on latency, escalations and cost, and get an alert when one is crossed.

aixplain One

How did booking-assistant do this morning?

Read 214 runs since 8:00
Checked them against your thresholds

212 answered on their own. One refund was escalated, one bulk cancellation was stopped, and one run crossed the 3 s latency threshold, so you got an alert.

Ask anything

MonitorToday · 214 runs

Time	Asked	Steps	Took	Cost	Outcome
09:41	Move Thursday to Friday	4	2.1 s	$0.004	Answered
09:38	Refund my 5-session package	3	1.4 s	$0.003	Escalated
09:30	Book Anna for Monday	5	3.4 s	$0.005	Answered
09:12	Cancel all of next week	2	0.9 s	$0.002	Stopped

Thresholds and alerts

LatencyAlert above 3 s1 run over · alert sent 09:30

EscalationsAlert above 5% of runs0.5% · OK

Cost per runAlert above $0.01$0.004 · OK

Analyze

See why runs end the way they do: the topics people raise, their sentiment, the error codes behind failures and the root cause of escalations.

aixplain One

Why are runs escalated?

Grouped this week's runs by topic and sentiment
Traced escalations to their cause

Most escalations are refunds over $200, which the refund policy caps. Calendar conflicts cause most errors.

Ask anything

AnalyticsThis week

Runs this week

Answered 95%Escalated 3%Stopped 2%

Topics

Rescheduling38%

Refunds27%

New bookings24%

Cancellations11%

Sentiment

Positive 71%Neutral 21%Negative 8%

Error codes

CAL_409Slot already taken12

PAY_402Card declined7

Root cause Refunds over $200 are escalated because the refund policy caps automatic refunds at $200

Evaluate

Run the same agent on two models over a dataset built from your real runs, and score both on correctness, conciseness, coherence and safety.

aixplain One

Compare booking-assistant on Claude Opus and GLM 5.3

Built a dataset from 1,200 of this week's runs
Ran the agent on both models
Scored both on 4 criteria

Claude Opus is more correct; GLM 5.3 is more concise and costs a quarter as much per run.

Ask anything

EvaluationModel A vs B

Dataset · booking-week-451,200 runs from this week, 42 marked by your reviewers

booking-assistant	AClaude Opus	BGLM 5.3
Correctness	86	81
Conciseness	78	84
Coherence	91	88
Safety	99	99
Cost per run	$0.004	$0.001

Scored by evaluators on correctness, conciseness, coherence and safety. Runs that both models miss are added to the dataset.

Evolve

aixplain One proposes a new version from the week's runs, with changed instructions, tools and inspectors, and scores it against the current one. You decide whether it ships.

aixplain One

Propose a better version from this week's runs

Rewrote the refund instructions
Added a policy lookup tool and a refund-limit inspector
Scored v4 against v3 on the same dataset

v4 scores 94 on correctness, up from 86. It ships only when you say so.

Ask anything

booking-assistantv3 → v4

	v3	v4 proposed
Instructions	Refunds over $200 are escalated	Check the refund policy first; escalate over $200 with the policy quoted
Model	Claude Opus	Claude Opus
Tools	calendar, payments	calendar, payments, policy lookup
Inspectors	Safety	Safety, Refund limit

Score	v3	v4
Correctness	86	94 ↑
Conciseness	78	83 ↑
Coherence	91	92 ↑
Safety	99	99

Ship v4Keep v3

What people get done

Streamline your toughest workflows, and reimagine the tools and apps you pay for. Pick one to try it.

InvoicesBooking appSales readoutCustomer answersInbox

“Build and deploy an agent that checks every invoice against its contract and reminds clients who haven't paid”

Reads each new invoice as it lands in Gmail
Checks it against the contract and your ledger
Sends a polite reminder when one is late, and tells you what it sent

Deployed on cloud, it runs every day at 9:00, with your computer off

Invoice follow-upLive on cloudEvery day at 9:00

Gmail

QuickBooks

ContractsToday 9:00 · 12 checked · 2.4s · $0.03

Client	Invoice	Amount	Due	Contract	Status
Brightwater Studio	0142	$1,800	14 days late	Matches	Reminder sent
Northline Dental	0139	$2,450	9 days late	Matches	Reminder sent
Okafor & Sons	0137	$960	21 days late	Rate differs	Needs you
Lumen Yoga	0144	$1,200	Due Friday	Matches	Waiting
Harbor Café	0131	$830	Paid Oct 20	Matches	Paid

To Sam at Brightwater · Invoice 0142Hi Sam, a friendly reminder that invoice 0142 for $1,800 was due on October 12. You can pay it from the link below. Thanks, Lina

“Build a booking app for my clients with an agent inside that answers their questions”

Plans the app and writes the code on your computer
Puts an agent inside that answers your clients' questions
Tests it and reviews the code with you

Deploy the app and its agent, with a key, a budget and monitoring

Booking appPreviewCodeTestsOn Desktop

Book with Lina · CoachingOctober

Tue 27Wed 28Thu 29Fri 30

9:0010:0011:3014:3016:0017:30

Coaching session60 min · video or in person$90

Intro call20 min · videoFree

Confirm Fri 14:30

Ask Lina's assistant

Can I move Thursday's session to Friday afternoon?

Yes. Friday 14:30 and 16:00 are free. Which would you like?

14:30 please

Done. Your session is now Friday at 14:30, and the new invite is in your inbox.

Deployed · booking-assistant v3Key, budget and monitoring onTests passed

“Every Monday, tell me where last week's sales came from”

Pulls last week's numbers from your CRM
Breaks revenue down by channel and flags what changed
Writes a short readout with what's worth a look

In your inbox every Monday at 8:00

Weekly sales readoutEvery Monday, 8:00

From HubSpot

Revenue last week$48,200+12% on the week before

New customers146 from referrals

Average deal$3,440+$210

Where it came from

Referrals$16,900+31%

Website$12,400+4%

LinkedIn$8,100−6%

Events$6,300+2%

Email$4,500−11%

Worth a look

Referrals closed three times faster than ads.

Two proposals over $10k have waited 8 days: Northline and Kestrel.

LinkedIn spend is up 20%, leads are flat.

Sent to you Mon 8:00

Posted in #salesNext readout Mon Nov 2

“Build an agent that answers our customers on WhatsApp from our docs”

Learns from your docs, Drive and Notion
Answers customers on WhatsApp and cites its source
Escalates what it can't answer to your team

Runs on cloud day and night, with every conversation on record

Customer answersLive on cloud

Hi, can I get a refund if I cancel my package after 2 sessions?

Yes. You get the unused sessions back, minus a $20 fee, within 5 working days. Shall I start it for you?Refund policy.pdf · page 2

And my gift card?

Lina handles gift cards, so I've escalated this to her. She'll reply today.Escalated to Lina

Today

Conversations46

Answered41

Escalated5

Learns from

Policies in Drive

Pricing in Notion

FAQ

“Every morning, sort my inbox and draft replies in my voice for me to approve”

Sorts what came in overnight
Drafts replies in your voice
Waits for you to approve each one, at your desk or from your phone

Every morning at 7:00

InboxSorted at 7:00

Gmail

Needs your reply · 3

Priya NairCan we move Thursday?

Sam OrtizContract renewal for 2027

Jonas WeberQuestion on invoice 0139

For your information · 9

CalendlyNew booking: Fri 10:00

Newsletters · 22 archived

Priya Nair · 6:52Hi Lina, something came up on Thursday. Could we move our session? Priya

Draft in your voice · to Priya

Hi Priya,

Thursday is tight for me, but Friday at 14:30 and 16:00 are both open. Which suits you better?

Lina

Approve and sendEdit

2 more drafts waiting. Approve them here or from your phone.

Explore the platform

What will you stop doing by hand?

Download aixplain One for Mac, Windows or Linux. Free to start, no card needed.

One agent, for every task

Hand it your work

Just ask

Works in your apps and files

Scan

Design

Athena picks the right model for each step

Start in aixplain Desktop, carry on in the CLI

Works with your files

Bring your own model

Stays on your computer

Review every change

Free to start, then pay as you grow

The agent factory

Develop and deploy

Monitor

Analyze

Evaluate

Evolve

What people get done

Explore the platform

What will you stop doing by hand?