Connectors & Tools

Give your agents access
to your business data

Connect Shopify, WooCommerce, Airtable, Supabase, Google Sheets — or any REST API. Agents read live data and take actions mid-conversation, so every reply is sourced from your actual systems, not guessed at.

Start free — 7 days Sign in
Built-in adapters

One-click connectors
for the tools you already use

Each built-in connector ships with a pre-wired tool pack. Connect your store and endpoints are auto-configured — no JSON editing, no docs to read.

Shopify

OAuth

Products, orders, customers, inventory, and shipping — via GraphQL and REST.

  • List & search products
  • Fetch order by ID
  • Customer lookup
  • Inventory levels
  • Shipping zones

WooCommerce

OAuth

Products, orders, and customers over the WooCommerce REST API.

  • List & search products
  • Fetch order details
  • Customer lookup
  • Shipping methods
  • Order status

Airtable

OAuth with PKCE

Browse your bases, read records, and search within any table.

  • List bases
  • List tables
  • Fetch records
  • Search within table
  • Filter by field

Google Sheets

OAuth with PKCE

List spreadsheets, read cell ranges, and search within sheets.

  • List spreadsheets
  • Read cell ranges
  • Search within sheet
  • Get sheet metadata
  • Named ranges

Supabase

Service role key

PostgREST queries against any table — reads and writes.

  • Query any table
  • Filter & search
  • Insert rows
  • Update rows
  • Delete rows

Any REST API

Generic connector

Paste an OpenAPI spec, a cURL command, or build endpoints manually. Supports all 5 auth modes.

Learn more
How it works

Tools fire
mid-conversation,
automatically

When a customer message matches a scenario, the agent checks which tools are in scope, calls them with AI-inferred parameters, and uses the live data to compose a precise reply. No hallucinating. No stale answers.

  1. 1

    Customer sends a message

    Arrives via WhatsApp or email.

  2. 2

    Scenario matches

    Agent identifies the intent and loads its tool set.

  3. 3

    Tool called

    e.g. shopify.get_order(#12345) — parameters inferred by Claude.

  4. 4

    Live data returned

    Status, tracking, ETA — fetched from the source.

  5. 5

    Reply composed

    Grounded in real data, sent in < 1 second.

tool trace · live

Inbound · WhatsApp

"Hi — did my order already ship? It was order #12345."

✓

Scenario matched: order-status

tools in scope: shopify.get_order

tool call

shopify.get_order("#12345")

→ status: shipped · 2026-05-22

→ carrier: USPS · ETA May 24

→ tracking: 9400 1110 2070 4837 5566

→ cache: MISS · fetched live

Reply drafted · 0.8s

"Your order shipped Thursday via USPS — expected Saturday. Tracking: 9400 1110 2070 4837 5566. Anything else?"

Generic REST connector

Any API.
Full power-user
control.

Not on the built-in list? Point at any HTTP endpoint. Paste a spec or a cURL, define your parameters, choose an auth mode — and you have a tool the agent can call.

5 auth modes

  • None

    Open endpoints — no credentials required.

  • API key header

    Key injected as a request header (e.g. X-API-Key).

  • Bearer token

    Standard Authorization: Bearer <token> flow.

  • HTTP Basic

    Username and password encoded in the Authorization header.

  • OAuth2 client credentials

    Server-to-server OAuth — client ID + secret exchanged for an access token automatically.

OpenAPI import

Paste a spec URL or upload a JSON/YAML file — endpoints are auto-generated. Schema, parameters, descriptions: all imported instantly.

# paste your spec URL

https://api.yourstore.com/openapi.json

# or upload JSON / YAML

→ 14 endpoints imported

cURL & param editor

Paste a cURL from DevTools and the endpoint editor auto-fills. Then define path, query, body, and header params with types, defaults, and descriptions.

# paste from DevTools

curl 'https://api.store.com/orders/12345' -H 'Authorization: Bearer tk_xxx'

→ method, URL, auth pre-filled

Response handling

Three ways to handle
what the API returns

Choose how each tool's response gets handed to the agent — from raw pass-through to fully canonicalized schema.

Raw Flexible

HTTP response returned as-is. Claude interprets the structure directly.

AI-interpreted Smart

Claude Haiku reads the response and extracts the relevant fields into a clean, structured shape.

Adapter-normalized Canonical

Dedicated adapter converts to a known schema — NormalizedProduct, NormalizedOrder, etc. Fastest for the main agent to consume.

Safety & control

Read-only by default.
Opt in to writes.

Every connector starts read-only. You choose when to give an agent write permissions — and which scenarios get access to which tools. The refund agent doesn't need the customer list. The after-hours agent doesn't need Stripe.

Write-tool toggle

Per-connector read/write switch. Merchants explicitly opt in before any AI can create, update, or delete records.

Per-scenario scoping

Choose exactly which tools each scenario can call. Principle of least privilege — applied to AI agents.

Operation labels

Every tool description is prefixed [READ], [WRITE], or [DELETE] so Claude knows the impact before calling it.

AI classification

Claude Haiku auto-classifies new endpoints as read, write, update, or delete when you import an OpenAPI spec.

Caching & performance

Smart caching.
Zero extra load
on your systems.

Two-layer cache — Redis for hot reads, Postgres for durable TTL records. Per-capability TTLs mean product lists cache for minutes while order statuses always fetch live. Writes automatically invalidate related reads.

Test-call mode supports a "force fresh fetch" bypass when you're debugging an endpoint.

Cache TTL examples per capability
Endpoint TTL Reason
Product lookup 5–10 min Warm cache on repeat questions.
Order status No cache Always live — customers expect accurate ETAs.
Shipping zones 1 hour Changes rarely; safe to cache aggressively.
shopify.get_order · last 7 days
24h7d30d
Error rate 0.4% last 7d
p50 latency 142ms median
p95 latency 390ms 95th pct
Cache hit rate 71% cost saved
AI cost estimate: $0.0012 / call avg
Analytics

Per-endpoint analytics.
See exactly
what's being called.

Error rate, p50 and p95 latency, cache hit rate, and estimated AI cost per endpoint. Switch between 24-hour, 7-day, and 30-day windows. Export any window as CSV.

Secrets vault

API keys encrypted
at rest. Never
exposed again.

Credentials are encrypted with AES-256-GCM (via Supabase Vault) immediately on save. They're never logged, never cached in plaintext, and never surfaced in the UI after the initial save. Not even we can read them.

  • AES-256-GCM encryption

    Industry-standard symmetric encryption at rest via Supabase Vault.

  • Never logged

    Request traces never capture credential values — only that auth was present.

  • Never cached

    Secrets are fetched per-request from the vault; no in-memory copies.

  • Hidden after save

    Once saved, the UI shows only a masked placeholder. No retrieval endpoint exists.

Plan limits

Tool caps by plan

More plan = more tools defined and more parallel calls per reply. Scale up as your use-cases grow.

Starter

16

tool definitions

3

calls per reply

Pro

32

tool definitions

6

calls per reply

Scale

64

tool definitions

10

calls per reply

Tool slots = total defined endpoint tools across all your connectors. Calls per reply = how many tool calls Claude can chain in a single response.

Get started

Connect your first tool
in under 5 minutes.

One-click OAuth for Shopify, WooCommerce, Airtable, and Google Sheets. Paste an OpenAPI spec for everything else. Your agent has live data before the first real customer message.

Start free — 7 days Sign in to your workspace

7-day free trial. Cancel anytime. Set up in 10 minutes.

Ready to build
your first agent?

7-day free trial. Connect your channels, write your first scenarios, and have your agent answering customers within the hour.

Start free trial Or talk to a human →