Start for free

Workflow engine for AI teams

Build it once. Watch every run.

Cubework strings your apps, models and people into workflows that run on their own, and keeps a readable record of every step they take.

Start for freeSee a live run

run_8f3a · lead scored 91 of 100 · 0.9s

Free for your first 1,000 runs. No card needed.

Teams at 600+ companies run on Cubework

  • Halcyon
  • Brightloom
  • Quarry
  • Fieldnote
  • Orbitly
  • Marrow

Running right now: live runs across 600+ teams

  • GBuild shipped, channel pinged3s ago
  • NBug logged, spec page opened7s ago
  • SCard declined, finance looped in12s ago
  • GBranch merged, release notes drafted19s ago
  • IChat marked urgent, ticket opened26s ago
  • SOrder paid, text message sent33s ago
  • HForm filled, lead created5s ago
  • NDoc changed, summary shared10s ago
  • WReport done, churn risk updated16s ago
  • SRefund cleared, payout sent22s ago
  • RReceipt odd, sent for review29s ago
  • ZReply written, customer answered38s ago

01 How a run works

Four steps, every single time.

When a run fails you see the exact step, the data it carried and the reason it stopped.

  1. 01

    Listen

    A webhook, a timer or an app event starts the run.

    • webhook
    • every 1m
  2. 02

    Gather

    Pull the context from your stack before anything is decided.

    • postgres
    • http get
  3. 03

    Judge

    Branch on rules, or let a model make the call inside guardrails.

    • if / else
    • model · tools
  4. 04

    Do

    Write back, ask a person, or hand off to the next workflow.

    • slack
    • retry x3
Median time per step
0.9s
Runs completed
99.97%
Ready connectors
380+

02 What is Cubework?

Cubework takes app events and turns them into runs you can read - each one passes through typed steps, model calls and human sign-off, so the work that used to wait in someone's inbox gets done on its own.

03 Connectors

Plug in anything. See every wire.

380 connectors, each with typed triggers and typed actions.

Sources06 connected

  • SSlack
  • SStripe
  • GGitHub
  • HHubSpot
  • NNotion
  • LLinear

Cubework core

  • OOpenAI
  • SSnowflake
  • SSalesforce
  • TTwilio
  • AAirtable
  • SShopify

Actions12 types

04 Recipes

Start from a recipe that already works.

Six of the 18 in the library. Copy one, swap the apps and ship it the same day.

  1. 01 · 4 steps

    Sort incoming support

    SslackMmodelLlinear

    slack.message > linear.issue.create

  2. 02 · 5 steps

    Score every new lead

    HhubspotMmodelSsalesforce

    hubspot.contact > salesforce.lead

  3. 03 · 3 steps

    Sum up sales calls

    NnotionMmodelSslack

    notion.page > slack.message.post

  4. 04 · 4 steps

    Catch odd payments

    SstripeMmodelSslack

    stripe.charge > slack.alert.send

  5. 05 · 4 steps

    Write the changelog

    GgithubMmodelNnotion

    github.release > notion.page.create

  6. 06 · 5 steps

    Answer refund asks

    SshopifyMmodelTtwilio

    shopify.order > twilio.sms.send

05 Teams

One engine, four different teams.

Pick a team and see the workflow it switches on in week one.

The outcome

No lead waits overnight.

Every form is enriched, scored and handed to the right rep before the visitor closes the tab.

6h to 80sfirst reply

The workflow

2.1s total

  1. Form sentHubSpot · 0.3s
  2. Data addedClearbit · 0.5s
  3. Scored 91 of 100Model · 0.9s
  4. Given to MinaSalesforce · 0.4s

The outcome

Urgent tickets jump the queue.

Messages are read, tagged and routed, and the angry ones reach a person within a minute.

3xfaster triage

The workflow

1.6s total

  1. Chat receivedIntercom · 0.2s
  2. Mood readModel · 0.7s
  3. Tagged urgentLinear · 0.3s
  4. Paged on callSlack · 0.4s

The outcome

Odd payments never slip by.

Each charge is checked against the customer's history, and anything strange is held for review.

0missed flags in Q3

The workflow

1.8s total

  1. Charge madeStripe · 0.2s
  2. History pulledPostgres · 0.4s
  3. Risk 78 of 100Model · 0.8s
  4. Held for OlaSlack · 0.4s

The outcome

Releases write their own notes.

Merged pull requests become a readable changelog, posted where the whole company can see it.

4hsaved per release

The workflow

2.4s total

  1. Release taggedGitHub · 0.3s
  2. Commits readGitHub · 0.6s
  3. Notes writtenModel · 1.1s
  4. Page postedNotion · 0.4s

06 Pricing

Pay per run. Seats cost nothing.

Slide to your real volume. The plan changes on its own, and the price you see is the price you pay.

Slide to set your volume

5kruns per month

You would pay

$69per month

One run = one workflow, from start to finish.

  • Starter1k includedFree
  • Team10k included$69 per month
  • Scale50k included$229 per month
  • EnterpriseVolume pricingLet's talk

07 Platform

Everything a run needs.

The parts your team touches on day one.

  • Any model, no lock-in

    Switch the provider per step. Your workflows stay the same.

  • Timers that never forget

    Schedules, webhooks or app events, with retries built in.

  • Every run, step by step

    Open the data at each step the moment something breaks.

  • Plug into what you run

    380 ready connectors, plus any HTTP endpoint.

08 Security

Hand your security team the log.

Real uptime history and the controls a security review asks for.

Uptime, last 90 days

99.97%

All systems running

90 days agoToday

  • SOC 2 Type IIAudited, report on request
  • GDPR and DPASigned on request
  • SSO and SAMLOn every plan, no add-on
  • Data residencyEU or US, your choice
  • Audit logRead-only, exportable
  • EncryptionAES-256 at rest, TLS 1.3

Your team stays the final step.

Copy a recipe, point it at your stack and let the runs begin. No card until you pass 1,000 runs.

Start for free