Conduit
Scheduled Sep 30, 2026Integration you run, on connectors you can read.
Replaces
- Zapier · Zapier, Inc. — $69 per org per month (Team (starting price)), about $828/yr at 400 seats
- Workato · Workato, Inc. — no published list price
Rented, per year
$828
Zapier list price at 400 seats.
Owned, per year
$8,280
AWS infrastructure only. Azure is $8,880.
Build
11 weeks
2 engineers. To parity on the ten core features, including the connector SDK and the first twelve production connectors. Each connector beyond that is roughly two to four engineer-days depending on the API's auth and pagination model.
Difficulty
4/5
Substantial. Compliance or realtime depth involved.
Reference organisation: 400 employees, 32 connected systems, 60 active workflows, ~8M step executions/month, 90-day run retention, single region, prod + non-prod. 3-year Compute Savings Plan on Fargate / 3-year reservation on Container Apps. Aurora minimum 0.5 ACU with non-prod stopped outside business hours. Run payloads over 256 KB offloaded to object storage. Excludes vendor support plans and egress above 1 TB/month. These figures are infrastructure only and exclude the build and the connector maintenance effort.
What it is
Conduit is an integration platform: connectors, triggers, a durable workflow engine, retries with backoff, dead-letter handling and per-run observability, all running inside your own account. It is aimed at organisations where integration has stopped being a convenience and become load-bearing — payroll to GL, CRM to billing, ticketing to on-call — and where a metered task counter now sits between the business and its own data. Conduit's tasks are free because the compute is yours. What you pay for is the connector work, once.
Why this one stops making sense
- Task metering prices the wrong thing. A retry costs you a task. A polling trigger that finds nothing costs you a task. You are billed for the platform's inefficiency as well as your own volume.
- Zapier Team starts at $69/month and climbs with the task tier; Workato does not publish a price at all. Neither number is something you can plan a three-year budget against.
- Integrations are where your business logic actually lives — the field mappings, the dedupe rules, the exception handling. Encoded in a vendor's visual builder, that logic is neither reviewable nor testable nor portable.
- Credentials for every system you own end up held by one third party. That is a single blast radius across finance, HR, CRM and infrastructure.
- When a source API changes, you wait for the vendor's connector update. With Conduit, the connector is a TypeScript module in your repo and the fix is a pull request.
What ships
- Durable workflow engine with checkpointed steps, resumable after crash or deploy
- Connector SDK: typed schemas, auth handling, pagination and rate-limit awareness in one module
- Triggers by webhook, schedule, database change and queue message
- Retry policies per step with exponential backoff, jitter and a real dead-letter queue
- Idempotency keys enforced at the engine, so a replayed webhook does not double-post
- Run-level observability: every step's input, output, duration and error, retained and searchable
- Secret handling through the cloud's own vault — no credential ever stored in the workflow record
- Data mapping with a typed expression language, version-controlled and diffable
- Replay and backfill: re-run a failed window against the original payloads
- Per-connection throttles so a downstream system with a 10 req/s limit is respected by construction
Data model
Stack
- frontend
- Next.js 15 App Router + Tailwind + TanStack Query + React Flow for the workflow canvas
- backend
- NestJS on Node 22 — a control-plane API plus a separate worker fleet running the execution engine
- database
- PostgreSQL 16 with partitioned run and step-execution tables, Row Level Security per tenant
- cache
- Redis 7 for per-connection token buckets, distributed locks and idempotency keys
- queue
- BullMQ for step dispatch and timers, with a Postgres-backed durable log as the source of truth
- auth
- OIDC against the customer IdP; workflow credentials held in the cloud vault, never in the database
- search
- PostgreSQL tsvector over run payloads and error messages, with a GIN index on the JSONB
- ai
- Optional field-mapping suggestion and error triage through a pluggable LLM gateway
Parity, honestly
What we match, and what we do not.
The rows marked No are the important ones. If one of them is the reason you bought Zapier, keep buying it — and we will tell you that on the first call rather than the fifth month.
| Capability | Zapier | Ours | Note |
|---|---|---|---|
| Webhook, schedule and polling triggers | Yes | Yes | — |
| Multi-step workflows with branching and loops | Yes | Yes | — |
| Retries, backoff and dead-letter handling | Yes | Yes | Ours are configurable per step rather than per workflow. |
| Per-run logs with full payload inspection | Partial | Yes | Zapier truncates and expires task history by plan. Conduit's retention is a Postgres partition policy you set. |
| Unmetered execution volume | No | Yes | Cost scales with compute, not with a task counter. |
| Replay and backfill against original payloads | Partial | Yes | — |
| Workflows as reviewable, testable code in git | No | Yes | Definitions are declarative files; the mapping expressions have unit tests. |
| Directory of 8,000+ pre-built app connectors | Yes | No | This is the honest gap. We build the 15 to 40 connectors you actually run and we build them well. If your integration estate is a long tail of hundreds of rarely-used SaaS tools, keep Zapier for the tail. |
| Vendor keeps connectors current when third-party APIs change | Yes | No | You inherit that maintenance. Budget a few engineer-days a quarter across a 30-connector estate. We hand over the test harness that tells you which one broke. |
| Non-technical business users building their own automations | Yes | Partial | Conduit ships a visual builder over the same definitions, but it is aimed at an ops analyst, not at anyone in the company. |
| Vendor-run SOC 2 report covering the integration layer | Yes | No | Conduit runs inside your existing control boundary and inherits your posture. If your auditor wants a third-party attestation on the iPaaS itself, that is a reason to keep buying one. |
| AI-assisted mapping suggestions from sample payloads | Partial | Yes | Runs against your LLM gateway on sample data you approve. |
| On-premise agent for systems behind the firewall | Yes | Yes | Conduit already runs in your VPC, so most of the reason for an agent disappears. |
Reference architecture
Built twice, on purpose.
Every application in Techtons ships with a production architecture for AWS and one for Azure, using the same diagram grammar so you can read them side by side. Your cloud, your account, your bill.
Amazon Web Services
- Compute
- ECS Fargate services (api, scheduler, worker) behind an ALB, workers scaled on SQS depth
- Data
- Aurora Serverless v2 PostgreSQL, ElastiCache Redis for rate limiting and locks
- Storage and edge
- S3 for large payload bodies and run archives, with a lifecycle rule to Glacier Instant Retrieval
- Identity
- Cognito federated to the customer IdP; connector credentials in Secrets Manager with rotation
- Observability
- OpenTelemetry to CloudWatch and X-Ray, one trace per run spanning every step
- Infrastructure as code
- Terraform, one module per environment, remote state in S3 with a DynamoDB lock table
Services
The decision worth arguing about
The non-obvious decision is not using Step Functions for orchestration, which is the reflex answer on AWS. Step Functions Standard bills $0.025 per 1,000 state transitions. The reference org runs about 8M workflow executions a month at roughly six steps each — call it 48M transitions, which is $1,200 a month in orchestration billing alone, more than the rest of the platform put together. So Conduit runs its own durable execution log in Postgres: each step commits its checkpoint in the same transaction as its output, and a worker crash resumes from the last committed step. The trade is that we now own the hard parts Step Functions gives you free — timer durability, exactly-once semantics at step boundaries, and the visual execution history — which is roughly two of the eleven build weeks. We take that trade because the alternative is a per-transition tax that grows with exactly the thing the platform is for. EventBridge Scheduler handles cron triggers at $1.00 per million invocations, which is cheap enough not to argue with.
Microsoft Azure
- Compute
- Azure Container Apps (api, scheduler, worker) with KEDA scaling on Service Bus queue length
- Data
- Azure Database for PostgreSQL Flexible Server, Azure Cache for Redis
- Storage and edge
- Blob Storage for payload bodies and archives, cool tier after 30 days
- Identity
- Microsoft Entra ID directly; connector credentials in Key Vault with managed-identity access
- Observability
- Azure Monitor and Application Insights, OpenTelemetry-native, one operation per run
- Infrastructure as code
- Bicep in an azd-compatible layout
Services
The decision worth arguing about
The Azure shape differs in one place that matters: ordering. Several connectors need per-connection FIFO — you cannot process the update to a Salesforce record before the create — and on Azure the natural answer is Service Bus sessions, keyed by connection id. The non-obvious decision is staying on the Standard tier rather than moving to Premium. Premium bills per messaging unit at roughly $0.93 an hour, about $670 a month for a single unit, which would roughly double the reference org's infrastructure bill for a platform whose steady-state message rate is nowhere near needing it. Standard supports sessions and bills per operation instead. The cost is that Standard has no guaranteed throughput and no VNet integration, so we shard session ids across a hash of connection plus entity to keep a single hot connector from serialising behind one session, and we accept public-endpoint Service Bus with Entra-only auth rather than private link. If the customer's network policy forbids that, Premium goes back in and the run cost rises by about $670 a month — we would rather say so than hide it in a diagram. We deliberately do not use Logic Apps Standard: it is a good product, but it is another vendor runtime, which is the thing we are removing.
Sources
Every price on this page, with the page we read it from and the date we read it.
| Product | Plan | List price | Checked | Source |
|---|---|---|---|---|
| Zapier | Team (starting price) | $69 / org/month | 2026-09-05 | zapier.com/pricing Zapier publishes Team as "Starting from $69/month". The figure is a floor, not a rate: the plan is metered by task and the price steps up with the task tier you select, up to 2M tasks/month. |
| Workato | — | Not published | 2026-09-05 | www.workato.com/pricing Workato does not publish list pricing. Its pricing page carries no numbers at all — only "Pricing built for flexibility" and a request-a-demo call to action. Every figure you will see quoted for Workato comes from a negotiated contract, so we do not publish one. |
Should you still be paying for Zapier?
Two weeks, fixed price. We audit what you actually use, map it against Conduit, cost the replacement on your own AWS or Azure account, and give you a delivery plan. If the honest answer is to keep the licence, that is what the report will say.