AI that makes complex automation, fast.

Turn your cross-system processes into reliable workflows through chat. Ballet builds the agents, writes the logic and integrations, and runs the work at scale.

Get a demo

When a deal closes, write the handover from the call and send it to the owner.

Get started with some examples

  1. Deal closes 1.6s
  2. Read the call 2.4s
  3. Draft the handover 2.1s
  4. Send to the owner 1.3s

Run complete in 7.4s

coassemble
smokeball
hirevue
pave
relevance
buffer
snaplogic
huntress

Automate work that spans four systems. No engineer required.

Complex knowledge work is never finished in one system. It runs through the CRM, the billing service, a database and a person — which is why it has stayed manual. Ballet is the first tool that automates it reliably.

Reach the service nobody documented.

The integration gets written as part of the build. That covers the CRM everyone has and the service your team wrote in 2019 that has no docs.

See the platform
Refund Request Resolution Connectors
  • Salesforce
    OAuth · 14 objects
    Connected
  • Zendesk
    API token
    Connected
  • Stripe
    Restricted key
    Connected
  • Billing service
    Internal · no public API
    Written by Ballet

Build like you have an engineer on call.

Ballet's AI coding agent can handle the kind of logic, integrations, and multi-system processes that usually need custom engineering — while you build it all through conversation.

See the platform
Refund Request Resolution Code
check-eligibility.js package.json
1
2
3
4
5
6
7
const deals = await sf.query(
  `SELECT Id, Amount FROM Opportunity`
);

for (const d of deals) {
  if (d.Amount > 50000) await gate(d);
}

Get the same result twice.

The model reads the ticket. Code moves the money. Run the same input twice and you get the same output, which is the part Finance asks about first.

Security & governance
Refund Request Resolution Play
  1. Read the request
  2. Check what's eligible
  3. Refund over $5,000 Waiting on Jess
  4. Post the outcome

Replay any run from six months ago.

Open a run from six months ago and step through it with the data it actually saw.

Security & governance
Refund Request Resolution Runs
  • ec56e7e3 Schedule
  • b1f0a92c Slack · Jess F
  • 7d24c118 Claude
  • 41a9e07f Chrome panel

Go from a sentence to a live playbook.

Describe the outcome and Ballet builds the steps, writes the integrations, and hands your engineers something they can read.

See who is using what, and what it costs

Every playbook, every run, every model — in one place.

Workspace usage 7d30d90d

US$2,603.62

24,048 runs 3,648 tool calls

PlaybookMemberModelTool

  • Slack Support Triage US$883.87 36%
  • RFP Response Drafter US$256.80 10%
  • Invoice Approval — Review Loop US$149.57 6%
  • Daily Prospect Call Prep US$70.92 3%

AI Gateway

Which models your builders can use.

  • anthropic/claude-opus-5 500K
  • openai/gpt-5.1-codex-max 400K
  • google/gemini-omni-flash 1M

292 models · 5 providers enabled

  • Spend by playbook, member, model or tool
  • Any run replayable months later
  • Approvals and model access set once, centrally

Pick the job you'd hand over first

Route each attendee while your event is still warm

Today A list lands after every event and someone spends two days matching it to the CRM, filling gaps, and deciding who follows up.

Event and webinar lists become routed, enriched leads before the follow-up goes cold.

  • Routed the same day
  • No manual matching
  • Rules in one place

All use cases

Marketing playbook Play
  1. Attendee list Trigger
  2. Match and enrich
  3. Apply the rules
  4. Route to an owner Done

Hand your account over the moment it closes

Today Closed Won fires and the handover depends on whether the rep had time to write one, and how much of the call they remembered.

Closed Won turns into a complete handoff instead of a note someone writes later.

  • Nothing dropped at handover
  • Written from the call
  • Owner notified

All use cases

Sales playbook Play
  1. Deal closes Trigger
  2. Read the calls
  3. Draft the handover Judgment
  4. Post to the owner Done

Keep your CRM clean without a cleanup project

Today Duplicates and missing owners pile up until someone books a cleanup project, and by the time it ships the data has drifted again.

Duplicates, missing fields and stale owners get fixed continuously, not in a quarterly project.

  • Fixed nightly
  • Safe merges only
  • Exceptions flagged

All use cases

RevOps playbook Play
  1. Nightly sweep Trigger
  2. Find the conflicts
  3. Merge what's safe
  4. Flag the rest For review

Give your agents a reply they only review

Today An agent reads the ticket, searches the docs, checks the account, and writes the same reply they wrote last week.

Common tickets arrive as a draft reply an agent reviews rather than writes.

  • Draft ready on arrival
  • Cites the docs
  • Agent still approves

All use cases

Support playbook Play
  1. Ticket arrives Trigger
  2. Search the docs
  3. Draft the reply Judgment
  4. Agent reviews Done

Chase your invoices the day they age

Today Invoices age quietly until someone runs the report, and the chase goes out without the account context that would settle it.

Overdue invoices get chased the day they age, with the context the customer needs.

  • Chased the day it ages
  • Context attached
  • Logged against the account

All use cases

Finance playbook Play
  1. Invoice ages Trigger
  2. Pull the history
  3. Write the chase Judgment
  4. Send and log it Done

Get AI's reasoning with a workflow's certainty.

Ballet AI agents Claude OpenAI Codex Dust Traditional workflows n8n Zapier Workato
Speed to build Minutes, plain English Fast to start Weeks, by hand
Connects to internal systems Reaches your proprietary systems Limited Not without weeks of ENG effort
Consistency Same output every time Different & unpredictable Consistent but rigid
Reasoning Flexible and considered Deep reasoning Inflexible, can't adapt
Auditable Yes, transparent code Minimal reporting Minimal reporting
Cost Predictable Expensive Climbs with volume

Everything else worth knowing

What is Ballet?

Ballet is a workflow-automation platform from the team behind Brainfish. Describe a multi-system outcome in plain English and Ballet builds it as reviewable, version-controlled code — deterministic where accuracy matters, agentic where judgment helps.

How is Ballet different from Zapier or n8n?

Connector libraries cap out when your stack is proprietary or the graph gets huge. Ballet writes the integration against any API the moment you need it, and the workflow runs as inspectable code instead of a brittle visual graph.

Is the output deterministic?

Yes where it matters. Ballet writes deterministic workflow code your engineers can read, approve, and replay. Agentic reasoning is used only on steps that benefit from flexibility — not on the steps that must never drift.

How does security and compliance work?

Generated code is visible and version-controlled, with approval gates and replayable runs. Ballet is part of Brainfish and covered by Brainfish’s SOC2 Type II compliance. Details live in the Trust Center at https://trust.brainfi.sh/.

Where does Brainfish fit?

Brainfish is the knowledge layer; Ballet is the execution layer. Customers can trigger Ballet workflows from Brainfish, and Ballet also runs standalone against your existing stack.

Bring the workflow that’s been in your backlog for two quarters.

We’ll model it on your real systems in a 30-minute session. No SOW, no scope creep.