ooligo

Temporal

workflow-automation durable-execution · agent-orchestration · developer-platform
API
RevOpsLegal OpsRecruiting & TACustomer Success
7.7 /10

What it is

Temporal is a durable-execution platform: you write a workflow as ordinary code in Go, Java, TypeScript, Python or .NET, and Temporal records every step so the workflow survives crashes, deploys, rate limits and week-long waits without losing its place. If a worker dies halfway through, another worker replays the history and carries on from the last completed step. Retries, timeouts, timers and signals from people or other systems are primitives, not code you write yourself.

The server is open source under the MIT licence and can be self-hosted. Temporal Cloud is the managed service, and it is where the company makes its money. Temporal says more than 4,000 customers run on it, naming OpenAI, Cursor, Netflix and JPMorgan Chase. It is also infrastructure other vendors build on: UiPath Maestro, UiPath’s agent orchestrator, runs on Temporal’s engine.

What it is not: a visual automation tool. There is no drag-and-drop canvas, no library of prebuilt SaaS connectors, and no form builder for business users. Every workflow is code that an engineer writes, tests and deploys.

Why ops leaders run into it now

Agent workflows are long-running by nature. An agent that researches an account, drafts a proposal, waits two days for a manager’s approval and then updates the CRM has to hold state through LLM timeouts, tool failures and a human pause. Temporal’s answer is that the agent loop becomes a workflow and every model call and tool call becomes an activity, so a failed call retries on its own and a completed one is never re-run and re-billed. Its integration with the OpenAI Agents SDK is generally available, and since Replay 2026 in May it has shipped Serverless Workers, in public preview on AWS Lambda since 3 August 2026, so teams no longer have to run their own worker fleet.

The company is well funded and independent: a $300M Series D in February 2026, then a $550M Series E at a $12.55B valuation on 14 September 2026, co-led by Lightspeed. In October 2026 it bought Oso, an authorisation startup, which points at permissions for agents as the next layer it wants to own.

A use case worth buying it for

An inbound-contract pipeline for a legal-ops team that handles 50,000 vendor agreements a year. Each contract starts a workflow: extract the text, have Claude classify risk clause by clause, route low-risk paper to auto-approval, and park anything non-standard on a signal that waits up to five days for a lawyer, with a timer that escalates after 48 hours. When the model provider rate-limits at 3 a.m., the activity retries with backoff instead of dropping the contract. When the team ships a new version of the classifier, in-flight contracts finish on the old code. That guarantee is the purchase.

Pricing

Temporal Cloud bills on Actions: each workflow start, activity start or retry, timer, signal, query and update counts as one. The entry plan has no base fee and charges $50 per million Actions plus support at 10% of usage; the old $100-a-month Essentials plan is gone from the price list. Business costs the greater of $500 a month or 10% of usage, includes 2.5 million Actions, and drops the overage rate in bands from $50 to $25 per million. Enterprise and the premium tier above it are annual contracts with 10 million Actions included and 24/7 support. Storage is billed separately per GB-hour. New accounts get $150 in credits valid for 90 days.

Real-world bands, at list price and before storage:

  • Pilot, 2M Actions a month (around 50,000 runs at 40 Actions each): about $110 a month on the entry plan.
  • Production team, 20M Actions a month: about $1,275 a month on Business.
  • Platform-scale, 100M Actions a month: about $3,850 a month on Business, before any committed-spend discount.

Self-hosting the open-source server costs nothing in licence fees, but you run the database, the visibility store and upgrades yourself. Budget at least one platform engineer’s time before it saves money.

Best for

Engineering-backed ops teams — a RevOps engineering pod, a legal-tech team, a platform group serving recruiting — that need agent or process workflows to run for hours or weeks with retries, human pauses and exactly-once side effects, and that want them in code under version control.

Alternatives, and when to pick them instead

  • AWS Step Functions — the default managed option for AWS teams, at $25 per million state transitions on Standard workflows. Pick it when everything already lives in AWS and the workflows are short state machines you are happy to define in JSON rather than code.
  • Azure Durable Functions — the default for Azure teams, with durable orchestrations written as code. Pick it when the team is on Azure and .NET and wants no new vendor.
  • Restate — the most visible new entrant, with a $20M Series A on 30 September 2026 and Replit Agent as a named user. Pick it when you want durable execution as a single lightweight binary and can accept a far smaller vendor.
  • Camunda or Appian — pick them when business stakeholders must read, approve or edit the process as a diagram, or when a regulator expects a BPMN model.
  • n8n — pick it when the workflow finishes in minutes, needs prebuilt connectors, and one ops person rather than an engineer will own it.

Watch-outs

  • Action counts surprise people. Retries, timers, signals, queries from the UI and each schedule run (three Actions) all bill, and a retry loop around a flaky API can multiply cost. Guard: count Actions per run in a staging namespace before committing to a plan, and cap retries on every activity that calls a paid API.
  • Determinism is a coding discipline. Workflow code must replay identically, so calling the clock, a random number or an API directly inside workflow code breaks it in production. Guard: keep every side effect in activities, add the SDK’s replay tests to CI, and use the versioning APIs before changing a running workflow.
  • It needs engineers, permanently. There is no ops-owned UI for building workflows. Guard: put Temporal only behind processes an engineering team already owns, and keep simple app-to-app automations on n8n or an iPaaS.
  • Serverless Workers are still in preview. Guard: run production workers on your own containers until Temporal marks the Lambda option generally available.