Mega

Your AI teammate for the Brainbase Conversational Platform.
Describe what you want. Mega builds it, deploys it, and manages it.

Docs · Based Reference · Examples

---

What is Mega?

Mega gives AI coding tools — Claude Code, Cursor, Codex, or anything that reads files — deep knowledge of the Brainbase platform. It's not a CLI, not an SDK, not an agent framework. It's a repo full of context that makes your AI tool an expert Brainbase engineer.

You say what you need. Mega knows the rest.

Write flows — from business requirements, PDFs, spreadsheets, or a conversation. Deploy anywhere — voice, chat, SMS, WhatsApp, API. One flow, many channels. Scale — templatize a working flow and deploy it across dozens of instances. Debug — pull logs, find where callers get stuck, fix the prompt, redeploy.

Quick start

git clone https://github.com/BrainbaseHQ/mega.git
cd mega
cp .env.example .env   # add your Brainbase API key

Open mega/ in your AI tool. It auto-configures:

Tool	Config	What Mega gives it
Claude Code	CLAUDE.md	Full platform context, Based reference, workflow patterns
Cursor	.cursorrules	Based syntax, platform conventions
Codex	AGENTS.md	Agent instructions for autonomous operation

Then just talk to it:

"Here's our customer support playbook [attach PDF].
 Write a Based flow for inbound voice calls that routes
 to sales, service, or parts based on caller intent."

"Deploy the booking agent to these 40 phone numbers
 [attach spreadsheet]. Use the dealership name and hours
 from each row."

"Pull logs from deploy_abc123. Where are callers getting
 stuck? Fix it and redeploy."

How it works — end-to-end

Here's what actually happens when you use Mega. This walks through building and deploying an appointment booking agent.

1. Describe what you need

"Build an appointment booking agent for a dental office.
 Greet callers, collect name and phone, let them pick a service
 (cleaning, filling, consultation), choose a time, and confirm."

2. Mega reads the Based reference

Your AI tool loads CLAUDE.md which points to docs/based.md — the complete Based language spec. It learns loop/until/talk, how say() and .ask() work, constraints, and patterns. It also finds examples/appointment-booking.based as a working template.

3. Mega writes the flow

say("Thanks for calling Riverside Dental! How can I help you?")

loop:
    res = talk("You are Sarah, the scheduling assistant...", True)
until "caller wants to book":
    info = res.ask(question="What is their name?", example={"name": "Jane"})
    # ... collect details, confirm, book
until "caller wants to cancel":
    # ... handle cancellation
until "caller is done":
    say("Have a great day!")
    end_call()

4. Mega creates a worker and uploads the flow

./scripts/bb.sh workers create --name "Riverside Dental" --description "Appointment booking"
./scripts/bb.sh flows create <worker_id> --name "Booking" --code-file booking.based

5. Test directly against the engine

curl "https://studio.brainbaselabs.com/v1/chat/completions\
?agent_id=<worker_id>&session_id=test-001" \
  -H "Authorization: Bearer <engine_key>" \
  -H "x-brainbase-api-key: <your_api_key>" \
  -H 'x-initial-state: {"variables": {}}' \
  -H "Content-Type: application/json" \
  -d '{"model":"<model>","messages":[{"role":"user","content":"I need a cleaning"}]}'

6. Deploy to a real channel

./scripts/bb.sh deployments create-voice <worker_id> \
  --flow-id <flow_id> --phone "+15551234567" --name "Main Line"

Scaling: one flow, many deployments

Write one flow with variables:

name = variables.get('office_name', 'our office')
hours = variables.get('hours', '9am-5pm')

Deploy it 40 times with different values per office. Tell Mega:

"Here's a spreadsheet with 40 offices — name, hours, phone,
 services. Deploy the booking flow to each one."

It reads the spreadsheet, creates workers, and deploys each with the right variables.

What's inside

mega/
├── docs/
│   ├── based.md           # Complete Based language reference
│   ├── platform.md        # Workers, flows, deployments, resources
│   └── deployments.md     # Deployment types and configuration
├── examples/              # Validated, engine-tested Based flows
│   ├── inbound-router.based
│   ├── appointment-booking.based
│   ├── order-taking.based
│   ├── lead-qualification.based
│   └── outbound-campaign.based
├── scripts/
│   └── bb.sh              # CLI wrapper for the Brainbase API
├── CLAUDE.md              # Claude Code context
├── .cursorrules           # Cursor context
└── AGENTS.md              # Codex context

The Based language

Based is Python + a small set of constructs for multi-turn, LLM-driven conversations. If you can write Python, you can write Based.

loop:
    res = talk("You are a helpful assistant.", False)
until "user wants to schedule":
    details = res.ask(question="When?", example={"date": "tomorrow", "time": "2pm"})
    say(f"Booked for {details['date']} at {details['time']}.")
until "user says goodbye":
    say("Take care!")

See docs/based.md for the full reference.

Links

• Brainbase docs
• Mega docs page
• Based Language Fundamentals

Contributing

Issues and pull requests welcome.

License

MIT