Loading…
Solutions · Integrations · Canarlo
Next.js 16, Supabase, TypeScript. For the moment the API stops being friendly — OAuth refresh, rate limits, signed payloads, replay. Typed contracts on every call, dead-letter queues, an audit row per attempt. The connector your CRM vendor calls a roadmap item.
Who this is for
Stripe to NetSuite. HubSpot to the data warehouse. The integration fails on Tuesday and nobody notices until the month-end reconciliation. We ship signed webhooks, dead-letter queues, and an alert that reaches a human before the auditor does.
Customers connect their Shopify, their Xero, their Gmail. OAuth refresh has to be invisible. Rate limits have to be respected. We ship the connector surface as part of the product, not an add-on that breaks under real traffic.
Forty-three Zaps, no documentation, two of them critical to billing. Half built by a contractor who left. We rewrite the integrations into typed TypeScript, observable per call, and you stop paying per-task pricing forever.
What we ship
Foundations
Operator accounts, admin accounts. Email, magic link, TOTP. Every login event logged with IP and user agent. Roles enforced at the database, not in the admin UI.
Customer A's OAuth tokens are unreachable from customer B's session. Tenant boundary enforced at the row. A misrouted query returns nothing — never another tenant's secret.
Every external call captured with target, payload hash, response status, retry count. The trail your compliance team needs the day a vendor disputes an invoice.
Structured JSON logs with secrets redacted. Sentry wired before launch. Per-integration latency, error rate, retry count live on a dashboard from day one.
Every pull request gets a live URL. Tests gate merges. Sandbox-mode credentials run on preview environments — production credentials never leave production.
Your Vercel team, your Supabase project, your domains, your keys. We deploy with your credentials and walk off at handover. No agency-held tokens.
Recent build
Illustrative scope
A typed integration layer between Stripe, HubSpot, and a warehouse. OAuth refresh handled, rate limits respected, signed webhooks both directions, dead-letter queue with replay. Six weeks, fixed fee. Same engineer answers the on-call page when an API deprecates next year.
Tech stack
Our process
Step 1
01
One scoping call. We map every integration — endpoint, auth flow, rate-limit policy, failure mode. Output: a written brief listing every contract. One week.
Step 2
02
Data model, queue topology, retry policy, idempotency strategy on the page. Replay surface designed. One week. You sign off before code is written.
Step 3
03
Four to eight weeks. Weekly demo on a real preview URL. Every external call audited from day one. Sandbox credentials before production tokens.
Step 4
04
Deploy to your cloud, cut over from the legacy Zap, watch the first week of real traffic. Handover doc names the failure mode and the on-call step.
Step 5
05
Optional retainer — vendor API changes, dependency updates, new connectors as your stack grows. From £500 a month. Same engineer. Cancel any time.
Parent service: Automations
Pricing
Fixed fee, scope written down before billing starts. £15k buys two integrations with OAuth, retry, and a monitoring dashboard. £30k buys five integrations with replay, dead-letter, and a customer-facing connector surface. £50k buys the integration platform — typed contracts, versioned schemas, alerting your on-call rotation respects.
Full pricing rationale and cost breakdown: How much does AI engineering cost?
Frequently asked
Anything with an API. Stripe, HubSpot, Salesforce, Pipedrive, Xero, QuickBooks, Slack, Linear, Notion, Airtable, Postgres, BigQuery, Snowflake. SOAP and XML when the partner is old enough to need it. If a vendor publishes docs, we wire to them. If they don't, we build the polling job or the inbound webhook handler.
Handled. OAuth flow per provider, refresh token rotation before expiry, encrypted storage at rest, audit log on every issuance. If a token revokes mid-run, the workflow pauses and routes a re-auth prompt to the right user. No silent failure, no token graveyard in someone's password manager.
Token-bucket per provider, per credential. Respects the 429 with the Retry-After header instead of guessing. Spikes get queued and drained at the allowed rate. The integration does not become the reason your vendor's compliance team calls. Cost-per-call surfaced in observability so a runaway loop isn't a billing surprise.
Dead-letter queue on every outbound integration, surfaced in the admin with the payload and the error. Replay from the UI once the upstream is healthy — idempotency keys mean a replay doesn't double-write. Inbound webhooks verified with HMAC and stored raw before processing. Nothing is lost to a transient outage.
Your infrastructure. Vercel and Supabase by default, self-host if you prefer. Credentials in your secret store, logs in your account, billing direct to you. No Canarlo SaaS in the request path. If we disappear tomorrow, the integration keeps running until something upstream breaks.
Start here
Twenty-minute call to map your endpoints. Proposal in your inbox inside forty-eight hours.