For an AI agent that runs 2,000 times a month, Temporal Cloud bills about $2, and self-hosted Temporal bills nothing for the software.

Neither number includes the two biggest costs: the workers you run, and the agent loop you write. A hosted agent loop like Cadenya has neither, for $49 a month. This page adds up all three.

Temporal’s prices come from temporal.io/pricing and Temporal’s docs, and Cadenya’s from cadenya.com/pricing, all checked on September 28, 2026.

What each option bills

	Temporal Cloud	Self-hosted Temporal	Cadenya
Base fee	None (“from $0/month”)	None: MIT license	$0 on Free, $49 a month on Growth
The unit	Actions: $50 per million for the first 5 million, then $45, $40, and lower	Your servers and database	Loops (one LLM request): 1,000 a month free, 50,000 on Growth, then $0.0015 each
Storage	$0.042 per GB-hour for open runs, $0.00105 per GB-hour for closed ones	Your database	Execution logs kept 14 days, included
Support	10% of usage	Community Slack and forum	Slack support on Growth
Credits for new accounts	$150, valid for 90 days	None	$25 of inference, once, at sign-up
What you run	Workers, which run your code	The server (four services), a database, and workers	Nothing. Cadenya calls your API for tool calls
What you write	The agent loop, as a workflow	The same	Tool sets, prompts, and approval rules
Model costs	Your model provider	Your model provider	Your model provider

Temporal Cloud pricing: Actions, and what counts as one

Temporal bills “Actions”, and a few kinds of them add up in an agent:

1. Starting a workflow.
2. Starting or retrying an activity. In an agent, every model call and every tool call is usually an activity.
3. Starting a timer, including the one Temporal starts for you when a wait has a timeout.
4. Every Signal, Query, and Update. Viewing a workflow’s call stack in the Cloud UI runs a Query.
5. Activity heartbeats that reach the server.

Replaying a workflow on your worker is free. Retries aren’t.

Example: 2,000 runs a month

Take a refund agent. Each run starts once, calls the model 6 times, calls 3 tools, and waits for one approval with a 7-day timeout. That’s 1 start, 9 activities, 1 Signal, and 1 timer: 12 Actions a run. Assume each run’s history is 200 KB (the model’s inputs and outputs are stored in it), the run stays open for a day, and closed runs are kept for 7 days.

	Temporal Cloud	Cadenya
Units	24,000 Actions	12,000 loops
Units billed	$1.20	Included in Growth
Storage for open runs	$0.40	Included
Storage for closed runs	$0.07	Included
Support	$0.17	Included
Monthly bill	about $1.84	$49

Cadenya’s free plan covers 1,000 loops, about 160 of these runs. Temporal’s $150 in credits cover the first 90 days.

The $100 and $200 floors you may have read

Some 2026 comparison posts quote a Temporal Cloud “Essentials” plan at $100 a month, or a basic tier at $200. Neither appears on temporal.io/pricing today. The page lists pay-as-you-go “from $0/month”, with “no base monthly fee”, and a Business plan “from $500/month” that includes 2.5 million Actions. If you don’t use Temporal Cloud in a month, “you will not be billed for it.”

What the bill leaves out

Workers. Temporal Cloud runs the server, not your code. Your workflows and activities run on workers you deploy, scale, and watch, on your own compute.

The agent loop. Temporal makes your code durable. The code is still yours: the loop that calls the model, the tool routing, the stream to your UI, context compaction, and the screen where someone approves. Temporal’s integration with the OpenAI Agents SDK (Python 3.10 or later) runs that loop inside a workflow, which helps. It’s still your code to run and upgrade.

The history cap. One run’s event history holds 51,200 events or 50 MB. A long agent conversation eventually starts a fresh run with Continue-As-New.

Self-hosted Temporal’s requirements and operational burden. The server is four services (Frontend, History, Matching, Worker) plus a persistence database: Postgres (PostgreSQL 13 to 16), MySQL 8.0.19 and later, or Cassandra. Self-hosted PostgreSQL can hold both stores, the persistence store and Visibility (searching runs, on PostgreSQL 12 and later from server 1.20), and Temporal recommends Elasticsearch for Visibility in production. The open-source bundle includes the Temporal UI and CLI. It supports the last three minor versions, so upgrades are part of the job, and so is being on call for it.

Which one costs you less

1. You already run Temporal: keep it for your workflows, and call Cadenya from an activity for each agent turn, so the loop, the approvals, and the streaming aren’t yours to build.
2. You run nothing yet: $49 buys the loop, the approvals, the streaming, and the chat widget, with no workers.

Other hosted loops bill by the hour instead, and the hosted agent runtimes comparison puts Claude Managed Agents, LangSmith Deployment, AWS AgentCore, OpenAI’s Agents API, and Cadenya side by side.

What does Cadenya offer that Temporal Cloud and self-hosted Temporal don’t?

Both Temporal options bill you for durable execution, and you build the agent on top. Cadenya’s $49 covers the agent itself: the loop, its durable state, human-in-the-loop approvals, and streaming, with no workers to run. Here’s what that covers:

1. No workers, even next to Temporal Cloud. Temporal Cloud runs the server, not your code, so you still deploy, scale, and watch workers. Self-hosted Temporal adds four services and a database. On Cadenya there’s no worker and no server, and your app stays a stateless web tier.
2. The agent loop, already written. Cadenya calls the model, runs the tools, streams each reply, and compacts the context window when a conversation gets long. On Temporal, that loop is a workflow you write, test, and version.
3. No versions to pin, patch, or drain. An objective keeps its variation’s configuration snapshot even when you edit the variation, and your deploys never touch a waiting run.
4. Human-in-the-loop approvals, screen included. A tool marked for approval pauses the objective on toolApprovalRequested until your backend, a webhook handler that posts to Slack, or a person in the React chat widget approves or denies it. On Temporal, the Signal, its handler, and the screen are yours to build.
5. Tools generated from your OpenAPI spec. A tool set turns every operation in an OpenAPI 3 document into a tool, re-checks the document each hour, filters the list, and loads tools on demand. MCP servers and plain HTTP endpoints work too.
6. Resumable streaming to your UI. The event stream honors Last-Event-ID, and the TypeScript SDK reconnects on its own. With Temporal, the stream to your UI is yours to build.
7. SDKs for the versions you run today. TypeScript on Node 18, Python 3.9, Ruby 3.1, and Go 1.22, plus a CLI and a plain HTTP API. Temporal’s Ruby SDK needs Ruby 3.2 at least, and its OpenAI Agents SDK integration needs Python 3.10.
8. One unit on the bill. There are no Actions, storage GB-hours, or support percentage to estimate. A loop is one LLM request, and waiting costs nothing.

Cadenya’s pricing page has the plan details, and the free plan needs no card.