For teams and their agents

Your AI agent Claude Code OpenAI builds automations. We run them.

Tell your coding agent what to build. Autobase runs it in the cloud — scheduled, shared, and auditable.

Join waitlist

For teams already building with AI agents.

Not another AI agent.

Not another builder to learn.

Autobase runs what your coding agent builds.

Intelligence once, automation forever

The agent thinks once. The work becomes real code that runs on a schedule, the same way every time, for pennies.

No developer knowledge needed

The agent handles everything technical: build, deploy, test, repair. Your team describes the job and stays in control.

Build. Your agent goes from a conversation to a deployed automation in one session.

app.autobase.so/workflows/invoice-reconciliation/overview
WorkflowsInvoice reconciliation

invoice-reconciliation

Invoice reconciliationAI

Status
Live
Visibility
Workspace
Live version
v1
Latest
v1
Creator
Maya Chen

What it does

Compares each morning's supplier invoice file against the ledger export, matches invoice numbers and amounts, asks AI to judge the unclear cases, and emails the mismatch report to the finance inbox.

When it runs

  • Every weekday at 07:30 (Europe/London)

Steps

  1. 1Fetch invoice filecodefetch_invoices
  2. 2Match against ledgercodematch_ledger
  3. 3Judge unclear matchesaijudge_matches
  4. 4Email mismatch reportemailsend_report

Latest version

v1 — Initial reconciliation build

claude — finance

> every morning I download the supplier invoice file and check it against our ledger export. can this run itself?

Yes. I'll build the reconciliation, test it on today's files, then deploy it to Autobase so it runs without your laptop.

Bash(npx autobase-cli@latest login)
⎿ Approved in browser

Created invoice-reconciliation · deployed v1
Dry run on today's files… green in 41s
⎿ 214 invoices matched · 3 mismatches flagged
Activated · scheduled weekdays 07:30 London

Done. It now runs every weekday morning and emails the mismatch report. Watch every run at app.autobase.so.

Run. The next morning it ran by itself. Nobody opened a laptop.

app.autobase.so/workflows/invoice-reconciliation/runs/run_01k2f8w7qj3m
WorkflowsInvoice reconciliationRuns

Invoice reconciliationAI

Succeeded

run_01k2f8w7qj3m

Purpose
Production
Started by
Scheduleweekday-mornings
Version
v1
Duration
26s
When
Today, 07:30
Model cost
$0.038 · 5,214 tokens · 1 call

Timeline

run26s
setup3s
fetch_invoices6s
match_ledger2s
judge_matches9s
send_report2s
teardown4s

Steps

1Succeededfetch_invoicescode6sdetails
2Succeededmatch_ledgercode2sdetails
3Succeededjudge_matchesai9sdetails
4Succeededsend_reportemail2s

Output

{
  "to": "ap-team@meridian.example",
  "delivery": "delivered"
}

Judge. AI only where judgement is needed, with a budget and a log. Everything else is plain code.

app.autobase.so/workflows/invoice-reconciliation/diagram
WorkflowsInvoice reconciliation

Invoice reconciliationAI

OverviewDiagramRuns2Versions1

Schedule

Weekdays 07:30

Fetch invoice file

Code

Match against ledger

Code

Judge unclear matches

AI

Email mismatch report

Email

judge_matchesAI
Model
Claude Sonnet
Instruction
Decide whether an unmatched invoice is a real mismatch or a formatting difference. Return a verdict with a reason.
This morning's run
$0.038 · 5,214 tokens · 9s

Oversee. Everything your team built, in one place. Not on laptops and shared drives.

app.autobase.so/workflows

Workflows

Filter workflows…
WorkflowVersionLast runCreator
Invoice reconciliationAIinvoice-reconciliation
v1LiveSucceeded2 hours agoMCMaya Chen
Inbox invoice intakeAIinbox-invoice-intake
v2LiveSucceeded26 minutes agoPNPriya Nair
Month-end accruals packmonth-end-accruals
v4LiveSucceeded5 hours agoMCMaya Chen
Weekly sales reportweekly-sales-report
v3LiveSucceeded4 days agoDKDan Kovacs
Supplier statement chaseAIsupplier-statement-chase
v2LiveSucceededyesterdayTOTom Okafor
Contract renewal reminderscontract-renewal-reminders
v1LiveSucceededyesterdayTOTom Okafor

Repair. It broke on a Tuesday. The agent that built it shipped the fix.

app.autobase.so/workflows/invoice-reconciliation/runs/run_01k5t2m9xq7v
WorkflowsInvoice reconciliationRuns

Invoice reconciliationAI

Failed

run_01k5t2m9xq7v

Purpose
Production
Started by
Scheduleweekday-mornings
Version
v1
Duration
8s
When
Today, 07:30

Run error

Workflow failure — the workflow, its configuration or one of its dependencies failed. The error below is the evidence a fix needs.

Error

{
  "code": "step_failed",
  "step": "fetch_invoices",
  "message": "Column 'Invoice No' not found in the
  supplier file. The layout may have changed."
}

Timeline

run8s
setup3s
fetch_invoices5s

Steps

1Failedfetch_invoicescode5sdetails
2Skippedmatch_ledger
3Skippedjudge_matches
4Skippedsend_report
claude — finance

> the reconciliation failed this morning, can you fix it?

Reading the run evidence from Autobase…
⎿ run_01k5t2m9xq7v · failed in fetch_invoices
⎿ Column 'Invoice No' not found. The supplier renamed it.

Patched fetch_invoices to accept both layouts
Deployed v2 · dry run on today's file… green
Activated · re-ran this morning's occurrence
⎿ 209 matched · 4 mismatches · report delivered

Fixed. Tomorrow runs on schedule as normal.

Automation or agent? There is no wrong answer. Different work wants different tools.

AutobaseClaude CodeOpenAI

Make it an automation

  • The execution is strict and cannot go wrong
  • The work repeats often and does not change
  • The volume is large and the cost must stay low
Invoice reconciliationDaily report packInbox file intake
Claude CodeOpenAI

Hand it to an agent

  • The task needs real judgement and creativity
  • The value comes from the agent's own input
  • The job is big, ambiguous or one of a kind
Market researchDrafting a proposalBuilding the automation

I've watched non-developers build brilliant automations with Claude Code, and I've watched every one of them die on a laptop. Autobase exists so the work your team builds actually keeps running: scheduled, supervised, and fixed by the same agent that built it.

Wayne Hsu

Founder

Questions.

What is Autobase?
The place where automations built by coding agents live and run. Your agent builds and maintains the automation, Autobase runs it in the cloud on a schedule, and your team oversees it with evidence for every run. Autobase is infrastructure, not another AI: it works with the agent you already use.
What can it automate?
Anything repetitive your team can describe: reconciliations, report packs, data pulls and checks, email chasing and reminders, file intake. Automations can also make HTTP calls to your own systems and tools, so they can read data out and write results back. If a coding agent can build it, Autobase can run it.
Which agents work with it?
Any coding agent. Claude Code, Codex, Cursor, OpenCode and whatever comes next: the API is plain REST with docs served to the agent itself, and the skill installs in one line. If your agent can call an API, it can use Autobase.
Do we need a developer?
No. The agent builds, deploys, tests and repairs. Your team describes the job, reviews the evidence and stays in control.
Is our data safe, and can we stop it?
Workspaces are isolated, AI calls run with zero data retention, and every workflow has a monthly AI spend cap. An admin stop switch blocks all new runs across the workspace, and every run leaves evidence you can audit.
What happens when something breaks?
You get an email that says what failed and whose fault it was, on Autobase's side or the workflow's. The run keeps its evidence, and the same agent that built the workflow reads it and ships the fix.

Give your team's automations
a place to run.

Join waitlist

For teams already building with AI agents.