Ship the logic. Skip the infrastructure.

Write a Python script and Aisle runs it on managed compute, with the schedule, secrets, retries, versions, and audit trail handled for you.

Start building Read the SDK docs

account-brief.py

Schedule · weekly

account-brief.py

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
# Find every open deal at this company.
company = aisle.inputs.get("company")
deals = aisle.integrations.pipedrive.search_deals(
    org_name=company,
)

# For each deal, grab news since the last run.
research = []
for deal in deals:
    news = aisle.ai.gemini_google_search(
        f"{company} news since {aisle.run.last_run_at}"
    )
    if news:
        research.append({"deal": deal, "news": news})

# Hand the lot to a saved prompt to write the brief.
brief = aisle.ai.run_prompt(
    slug="account-brief",
    research=research,
)

aisle.integrations.slack.send_message(
    channel="#revenue",
    text=brief,
)

Run · success4.2s · just now3 deals · 2 with news

Build, deploy and run deterministic AI workflows in Python.

Complete with integrations, RAG, smarter retries, versioning and more.

Run AI Prompts

In Aisle, prompts are versioned entities with their model settings and revision logs attached. Call any by slug.

python

aisle.ai.run_prompt("classify-inbound", {
    "subject": email["subject"],
    "body": email["body"],
})

Call Integrations

Call integrations like Slack, Telegram, Asana, Jira and more. Each one is namespaced under aisle.integrations.

python

aisle.integrations.slack.create_message(
    channel="#sales-inbound",
    text="New lead from Acme",
)

Search Memory

Search and store agentic memories, or query an on-the-fly RAG database, with a single call.

python

aisle.memories.search(
    "Acme renewal",
    folder="crm-notes",
)

Run in Parallel

Fan a function over a list of items. Concurrency, retries, rate limits, and checkpoints are arguments.

python

aisle.parallel(
    handle, items=new_emails,
    concurrency=5,
    retry=aisle.RetryPolicy(max_attempts=3),
)

Track Revisions

Every save snapshots the script, input schema, and entrypoint together. Restore any from history.

Diff Versions

Compare any two revisions of a task line-by-line. Reviewable like a pull request.

Read the SDK docs

Trigger a task from anywhere.

Run on demand, on a schedule, on an HTTP request, or in response to an event from any connected integration.

Manual / On-demandrun by hand

Schedulecron

APIHTTP endpoint

Bugsnagerror events

GitHubpush, PR, release

Gmailnew email

Google Newsnews topics

Jiraissue, comment

Microsoft Teamschannel message

Outlook Mailnew email

SERPsearch results

Slackmessage, mention

Telegrambot message

Webhookany HTTP POST

See all integrationsalso callable from any task

Distributing a task.

A task can hand its output to a person, return it to the calling system, write it into your knowledge base, or forward it to another service. You can also share the task itself or reuse it as a tool.

Output chat

Output a read-only or interactive chat, seeded with knowledge and a starting message so the recipient can ask follow-ups.

Sharing

Share the task with specific people or your whole organization.

Memories

Save the output to a knowledge folder, where Chat, Prompts, Projects, and connected tools can read it.

Project tool

The task becomes a tool a project agent can call during a chat.

Return value

The run sets a JSON result the caller can read.

Triggers

Other systems start the task through an API call, a webhook, or an inbound email or message.

curl + json

$ curl -X POST https://api.aisle.sh/tasks/account-brief \
    -H "Authorization: Bearer ak_•••••••••••" \
    -d '{"company": "Acme"}'

{
  "status": "success",
  "duration_ms": 4243,
  "result": {
    "company": "Acme",
    "risk": "high",
    "summary": "Acme is at high risk this quarter…"
  }
}

Easier to maintain than a node graph.

A canvas is hard to manage. Changes are onerous and tough to follow. A task moves the problem into code, where logic is easier to reason about, faster to modify, and version tracking is fast and simple.

account-brief.workflow

account-brief.py

# Find every open deal at this company.
company = aisle.inputs.get("company")
deals = aisle.integrations.pipedrive.search_deals(
    org_name=company,
)

# For each deal, grab news since the last run.
research = []
for deal in deals:
    news = aisle.ai.gemini_google_search(
        f"{company} news since {aisle.run.last_run_at}"
    )
    if news:
        research.append({"deal": deal, "news": news})

# Hand the lot to a saved prompt to write the brief.
brief = aisle.ai.run_prompt(
    slug="account-brief",
    research=research,
)

aisle.integrations.slack.send_message(
    channel="#revenue",
    text=brief,
)

Write your first task in the editor.

Free plan includes the runtime, integrations, and revision history.

Start building Read the SDK docs

Triage inbound sales emails every 15 minutes.

An example of a complete task: It runs on a schedule, classifies new email with a saved prompt, checks prior context in memory, and posts qualified leads to Slack.

triage-inbound-sales.py

*/15 * * * *

1def handle(email):
2    # Skip emails handled on a previous run (cache persists between runs)
3    if aisle.cache.get(f"seen-{email['id']}"):
4        return
5 
6    classification = aisle.ai.run_prompt("classify-inbound", {
7        "subject": email["subject"],
8        "body": email["body"],
9    })
10 
11    if classification["category"] == "qualified-lead":
12        notes = aisle.memories.search(email["from"], folder="crm-notes")
13        aisle.memories.store(
14            name=email["from"],
15            content=email["body"],
16            folder="crm-notes",
17        )
18 
19    aisle.cache.set(f"seen-{email['id']}", True)
20 
21new_emails = aisle.integrations.gmail.search_messages(
22    query="label:inbox is:unread newer_than:1d",
23    max_results=50,
24)
25 
26aisle.parallel(
27    handle,
28    items=new_emails,
29    concurrency=5,
30    retry=aisle.RetryPolicy(max_attempts=3, initial_backoff_ms=1000),
31)

1
No imports
No `pip install`, no requirements file. It is just the work.
2
No credential code
Gmail connects once on the task. Auth never appears in the script.
3
Prompts by slug
"classify-inbound" lives somewhere a non-engineer can edit.
4
Memory in one line
No vector database, no embedding model to pick.
5
Retries as a parameter
Concurrency, backoff, and rate limits in one place, not a task queue.

Tasks share state, prompts, and memory with the other building blocks. Reach for the shape that fits.

Write your first task.

Open the editor, write a script, and run it against your connected accounts.

Start building

Read the SDK docs.

Ship the logic. Skip the infrastructure.

Build, deploy and run deterministic AI workflows in Python.

Run AI Prompts

Call Integrations

Search Memory

Run in Parallel

Track Revisions

Diff Versions

Trigger a task from anywhere.

Distributing a task.

Output chat

Sharing

Connectors

HTTP / webhook

Memories

Project tool

Return value

Triggers

Easier to maintain than a node graph.

Write your first task in the editor.

Triage inbound sales emails every 15 minutes.

Projects

Memories

Reusable Prompts

Write your first task.

Ship the logic. Skip the infrastructure.

Build, deploy and run deterministic AI workflows in Python.

Run AI Prompts

Call Integrations

Search Memory

Run in Parallel

Track Revisions

Diff Versions

Trigger a task from anywhere.

Distributing a task.

Output chat

Sharing

Easier to maintain than a node graph.

Write your first task in the editor.

Triage inbound sales emails every 15 minutes.

Tasks alongside the rest of Aisle

Projects

Memories

Reusable Prompts

Write your first task.