Temporal
Also known as: Temporal.io, Temporal Cloud, Temporal Technologies
Open source durable execution platform whose Workflows as code survive crashes and run long-lived AI agents, self-hosted under MIT or on multi-region Temporal Cloud.
Temporal is an open source durable execution platform. Workflows written as code in seven SDK languages orchestrate Activities with retry policies and timeouts, Child Workflows, Signals, Updates, Queries, durable Timers and Schedules, and every step is recorded in an event history so a Workflow survives crashes and restarts. For AI work, its integrations run agents built on the OpenAI Agents SDK, Google ADK, LangGraph, Pydantic AI, Strands, Mastra and the Vercel AI SDK as Workflows with model and tool calls as Activities, and an AI cookbook covers agentic loops, human-in-the-loop, guardrails and multi-agent orchestration.
The server is MIT licensed and self-hosts in the customer's own infrastructure. Temporal Cloud runs across AWS and Google Cloud regions with multi-region replication, audit logs streamed to Kinesis or Pub/Sub, history export to S3 or GCS, user roles, SAML SSO and SCIM, and is SOC 2 Type II certified and GDPR and HIPAA compliant. Cloud pricing is pay-as-you-go from $0 a month with $150 in free credits, billed on Actions from $50 per million and on storage, with Business from $500 a month and Enterprise and Mission Critical through sales.
Vendor details
Canonical URL
https://temporal.io
Category
Agent infrastructure
Company status
independent
Use cases & customers
Target customers
Deployment options
In practice
Your research agent runs for three days across fifty documents and the server restarts mid task. Event history replay resumes from the last recorded step instead of repeating API calls and duplicating writes.
You already use the OpenAI Agents SDK, Pydantic AI or AWS Strands and want the agent to survive failures without rewriting it. Temporal has durable execution integrations for each instead of asking you to adopt its own agent framework.
Data residency rules prevent using a managed orchestration service. The open source server self hosts on your own infrastructure, so workflow data stays in your environment.
Sources & related URLs
Agentic Index coverage score
8.5 / 14 capabilities · 61%
| Integrations & Tool Calling | Partial |
|---|---|
|
Tools come from the customer's agent framework, which Temporal makes durable. The integrations catalog covers the Vercel AI SDK, Google ADK, LangGraph, LangChain Deep Agents, Mastra, the OpenAI Agents SDK, Strands and Pydantic AI, each running the agent as Workflows with model and tool calls as Activities, and any Activity can call an outside system. The tools, credentials and connectors belong to the framework and the customer's code. Temporal has no connector catalog or authenticated action framework of its own. Sourcedocs.temporal.io/integrationsread 2026-09-23 |
|
| Workflow Orchestration | Full |
|
Workflows written as code in seven SDK languages orchestrate Activities with retry policies and timeouts, Child Workflows, Signals, Updates and Queries, durable Timers, Schedules and Nexus calls across namespaces. Temporal is a workflow engine, and its AI cookbook runs multi agent orchestration (Google ADK) as Workflows. Sourcedocs.temporal.io/workflowsread 2026-09-23 |
|
| Knowledge Grounding & RAG | Not documented |
|
There is no retrieval structure that Temporal maintains over the customer's corpus. Its AI cookbook runs agent loops, deep research and tool calling as Workflows, but indexing and retrieval belong to the customer's code or framework. The Temporal Docs MCP server searches Temporal's own documentation, not the customer's. Sourcedocs.temporal.io/ai/cookbookread 2026-09-23 |
|
| Human Oversight & Guardrails | Partial |
|
Durable waits that Temporal ships let the customer's code hold for a person. Signals and Updates deliver a person's decision to a running Workflow, and durable Timers bound how long it waits. The AI cookbook's human in the loop agent and its post LLM guardrail with hard rule overrides show the pattern. The approval logic, approver identity and guardrail rules live in the customer's code, and Temporal has no approval step or review surface of its own. Sourcedocs.temporal.io/ai/cookbook/human-in-the-loop-pythonread 2026-09-23 |
|
| Security, Identity & Governance | Full |
|
Temporal Cloud is SOC 2 Type II certified and compliant with GDPR and HIPAA, with a Trust Center and a GDPR-aligned DPA. For access control, every Cloud plan includes user roles, service accounts, API keys, audit logging and SSO via SAML, with SCIM on Business and Enterprise, and the Cloud docs cover roles and permissions and custom roles. Encryption the customer holds, through a Codec Server, means Temporal cannot read payloads. Sourcetemporal.io/securityread 2026-09-23 |
|
| Observability & Auditability | Full |
|
Each Workflow is recorded step by step in its event history, browsable in the Web UI and queryable through visibility. Temporal Cloud streams audit logs to AWS Kinesis or GCP Pub/Sub, exports Workflow event histories to S3 or GCS, and exposes metrics that the integrations catalog sends to Datadog, Grafana, Elastic and ClickStack, with LangSmith and Langfuse tracing LLM calls inside Workflows. Retained storage is billed and sized by plan. Sourcedocs.temporal.io/cloud/audit-logsread 2026-09-23 |
|
| Memory & State Persistence | Partial |
|
Every step lands in the Workflow's event history, so after a failure the Workflow's state is rebuilt by replay and the code carries on from the restored outputs. That is real state, but it lets the same run finish and carries nothing into the next run. The history is browsable in the Web UI. Temporal's AI cookbook and integrations present no keyed state store across runs, and no long lived Workflow for each entity, as agent memory. Sourcedocs.temporal.io/temporalread 2026-09-23 |
|
| Deployment & Data Residency | Full |
|
Customers can self host or choose a region. The open source server self hosts in the customer's own infrastructure under a self hosted guide, and Temporal Cloud publishes its region list across AWS and Google Cloud with multi region and multi cloud replication and migration between regions. Sourcedocs.temporal.io/self-hosted-guideread 2026-09-23 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
Ready made starting points ship with Temporal, among them an AI cookbook of runnable examples (agentic loops with Claude and OpenAI, deep research, a durable MCP server, human in the loop, multi agent orchestration with Google ADK, agents on the OpenAI Agents SDK, Strands and the Vercel AI SDK), sample repositories across its SDKs, and agent skills and a Claude Code plugin for coding agents. These are examples a developer builds from, not a catalog of agents a customer adopts. Sourcedocs.temporal.io/ai/cookbookread 2026-09-23 |
|
| Triggers & Channel Coverage | Full |
|
Work wakes without a caller. Schedules start Workflow Executions on calendar or interval specs, and running Workflows receive Signals and Updates from outside systems and resume on durable Timers. Sourcedocs.temporal.io/scheduleread 2026-09-23 |
|
| Model Flexibility & Routing | Not documented |
|
No model runs inside Temporal for the customer to choose. Model calls are Activities in the customer's code or framework (the cookbook calls Claude, OpenAI and Gemini, and one recipe routes through LiteLLM, which is the customer's library), and Temporal records them whatever the model is. Sourcedocs.temporal.io/ai/cookbook/hello-world-litellm-pythonread 2026-09-23 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Official SDKs in seven languages, the Temporal CLI, the server's gRPC API, and service accounts and API keys for Temporal Cloud make the platform callable from outside. The Temporal Docs MCP server searches documentation only. Sourcedocs.temporal.io/encyclopedia/architecture/temporal-sdksread 2026-09-23 |
|
| Testing, Debugging & Optimization | Partial |
|
A testing suite covers the customer's Workflow code, with a test environment with time skipping, mocked Activities, and replay testing that runs new code against recorded event histories to catch nondeterministic changes. That tests whether a code change is safe to deploy, not how well an agent performs, and replay based debugging is also available. Temporal has no evaluation harness, scored test cases or quality gate on agent output of its own, and Braintrust in the integrations catalog is a third party. Sourcedocs.temporal.io/develop/python/testing-suiteread 2026-09-23 |
|
| Browser & Computer Use | Not documented |
|
The agent operates no browser or desktop, and Temporal has no browser automation environment of its own. It runs the customer's Workflow and Activity code on the customer's workers. Sourcedocs.temporal.io/ai/cookbookread 2026-09-23 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Temporal Proxy, now in public preview, is a component customers run in their own environment that encrypts Workflow payloads before they reach Temporal Cloud. It works with AWS KMS, Google Cloud KMS, Azure Key Vault or a custom key service, so Temporal Cloud stores only ciphertext it cannot read. Making the proxy the required connection to Temporal Cloud means no application or Worker can skip encryption by mistake.
Bears on: Deployment / data residency
View sourceServerless Workers for Google Cloud Run are now in public preview. Customers deploy Worker code to a Cloud Run Worker Pool in their own Google Cloud project, and Temporal watches the Task Queue, adds instances when work builds up and scales back to zero when the queue is empty.
Bears on: Workflow orchestration
View sourceTemporal's Python SDK 1.34.0 extends its experimental Google ADK integration to ADK v2 graph workflows and durable human approval steps, and runs AWS Strands sandboxes inside durable workflows.
Bears on: Human approval / guardrails
View sourcePricing
Pay-as-you-go from $0/mo plus usage · Business from $500/mo · Actions from $50 per million
Consumption: Actions (billable operations, per million, tiered) plus Storage (active and retained, per GB-hour), plus a support cost or plan fee; the open source server is free to self host
What is public
Plan structure, rates and the Action based consumption model are published on Temporal's pricing page.
Billing mechanics
Two entirely different cost models. Self hosting is free of license cost but carries the infrastructure and operational burden of running the Temporal server with Cassandra or PostgreSQL and Elasticsearch, modeled by third parties at roughly $2,500 to $4,500 a month at moderate scale. Temporal Cloud combines a monthly plan fee with consumption on Actions and on active and retained storage.
Cost watchouts
Every Workflow operation (starts, Activities, Signals, Timers, heartbeats) is a billable Action, and export and Nexus calls add Actions too; pay-as-you-go support costs 10% of usage, and Business charges the greater of $500 or 10% of consumption.
Variable cost rationale
Billing is consumption of Actions and storage with tiered rates, so Workflow design (how many Activities, Signals and Timers) and retention drive nearly all cost.
Additional watchouts
Independent guidance advises caution for teams that cannot yet estimate Actions or storage, which is precisely the position of anyone with unpredictable agent behavior. Learning curve and operational weight are the two most cited objections in comparisons against lighter alternatives.
Overage / add-ons
Consumption billed on Actions at roughly $50 per million, plus active and retained storage, on top of the monthly plan fee
Sales call required
Mixed (some tiers require a call)
Free / trial
$150 in free credits for 90 days on Temporal Cloud; the open source server is free to self host
Lowest paid plan
Business from $500 a month (or 10% of consumption)
Commercial notes
The Action is the unit that catches agent teams out. A background agent is not one run: starting a workflow, calling tools through activities, retrying failures, recording heartbeats, receiving signals and updates, setting timers and holding event history open all register separately. Comparative modeling puts Temporal Cloud above AWS Step Functions at equivalent volume, with durability rather than unit price as the justification.
Key ambiguities
Cost depends on how many Actions a Workflow design generates and how long histories are retained, which varies widely by workload; Enterprise and Mission Critical are annual contracts through sales.
Missing data
Enterprise and Mission Critical pricing, which is set through annual contracts with sales.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Temporal
The closest documented capability profiles to Temporal among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- DBOS8.5 / 14Matches Temporal across all 14 documented capabilities
- Hatchet8.5 / 14Matches Temporal across all 14 documented capabilities
- Restate8.5 / 14Fuller documented coverage on Memory & State Persistence
- Inngest10.0 / 14Adds documented Model Flexibility & Routing
- Bird7.5 / 14Fuller documented coverage on Integrations & Tool Calling
- C TWO8.5 / 14Fuller documented coverage on Integrations & Tool Calling and Human Oversight & Guardrails
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded