Restate
Also known as: Restate.dev, Restate Cloud, Virtual Objects
Code first durable runtime for agents shipping as a single binary: every step journaled for crash recovery, Virtual Objects for per session state, approval pauses lasting months, and rollback of in flight executions.
Restate is a code first runtime for durable services, workflows and AI agents, and its distinguishing choice is operational simplicity. It ships as a single binary with no extra databases and no separate worker processes, which independent comparison calls the least operationally demanding self hosted option in the durable execution category.
The mechanism is journaling. Every step an agent takes is written to a journal, so when a worker crashes or restarts, Restate replays the journal and continues from saved progress rather than repeating the whole workflow. On top of that sit Virtual Objects, stateful entities keyed by user or conversation with built in state and concurrency control, which give the platform a clean answer to per session agent state that most competitors leave to the developer.
Three capabilities stand out for agent work specifically. Human in the loop pauses can wait for approvals taking minutes or months and survive crashes in between. Multi agent orchestration runs over durable RPC with fan out and timeouts across agents, tools and services. And task control lets an operator cancel, kill, roll back or restart executions, individually or in bulk, through a UI or an API.
Restate also turns idempotency on by default, which most platforms leave to the developer, though the keys do not deduplicate across separate requests.
The honest trade off is stated plainly by independent reviewers: the team must adopt the Restate service and virtual object model, which is a real shift if the agents in question are plain Python loops.
Vendor details
Canonical URL
https://restate.dev
Category
Agent infrastructure
Subcategory
Durable execution runtime for agents
Funding status
Private. Funding detail not retrieved in this pass.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Designed to sit close to existing backend services rather than replace them, with durable RPC, fan out and timeouts across agents, tools and services. A named integration with Pydantic AI adds durable execution to Pydantic agents in a few lines, and observability pairs with Logfire for traces, evaluations, cost and behaviour in one view. Task control is exposed through both a UI and an API.
In practice
An agent crashes mid run and resumes from the journal on another process rather than repeating every completed step.
A workflow pauses for a human approval that takes six weeks, and the pause survives every restart in between.
An operator discovers a misconfigured agent has been acting incorrectly and rolls back the affected executions in bulk from the UI.
Sources & related URLs
Agentic Index coverage score
8.5 / 14 capabilities · 61%
| Integrations & Tool CallingDesigned to sit close to existing backend services rather than replace them, with durable RPC across agents, tools and services, a named Pydantic AI integration adding durability in a few lines of code, and Logfire pairing for traces, evaluations, cost and behaviour. Task control exposed via both UI and API. Pydantic documentation and independent comparison 2026-08-08 | Full |
|---|---|
| Workflow OrchestrationMulti agent orchestration over DURABLE RPC with fan out and timeouts across agents, tools and services, on a runtime built for durable services and workflows; every step is journaled so a crashed worker resumes from saved progress rather than repeating the workflow. Pydantic first party integration documentation and independent comparison 2026-08-08 | Full |
| Knowledge Grounding & RAGRestate is a runtime and holds no knowledge layer, retrieval or grounding capability; agents bring their own. Architecturally expected rather than a gap, and consistent with how other durable execution vendors are graded. Independent comparison 2026-08-08 | Unable to verify |
| Human Oversight & GuardrailsHUMAN IN THE LOOP IS A NAMED PRIMITIVE: pause for approvals that take MINUTES OR MONTHS, SURVIVING CRASHES IN BETWEEN. Paired with TASK CONTROL - cancel, kill, ROLL BACK or restart executions, individually or IN BULK, via UI or API. Rollback of in flight work is rare; most platforms let an operator stop an agent rather than undo it. Pydantic first party integration documentation 2026-08-08 | Full |
| Security, Identity & GovernanceNOT RETRIEVED - FLAGGED FOR VERIFICATION. Self hosting keeps data inside the customer boundary, which is a genuine control, but no trust centre, attestation, access control model or compliance evidence was retrieved in this pass. Check restate.dev security pages and correct. Independent sources 2026-08-08 | Partial |
| Observability & AuditabilityThe execution journal records every step and is what makes replay possible, so a complete record of what an agent did exists by construction, and a UI plus API expose running executions for inspection and control. Observability tooling proper is delegated to Logfire in the documented Pydantic pairing rather than provided natively. Pydantic documentation and independent comparison 2026-08-08 | Partial |
| Memory & State PersistenceDURABLE SESSIONS as a first class primitive: VIRTUAL OBJECTS are stateful entities KEYED BY USER OR CONVERSATION with built in state AND concurrency control, and state persists across executions. Independent comparison specifically credits this as a clean answer to per session state and concurrency that most platforms leave to the developer. Memory as a runtime primitive rather than a data store side effect. Pydantic documentation and independent comparison 2026-08-08 | Full |
| Deployment & Data ResidencyOPEN SOURCE AND EASY TO SELF HOST AS A SINGLE BINARY with NO EXTRA DATABASES AND NO SEPARATE WORKER PROCESSES - independent comparison calls it the least operationally demanding self hosted option in the category. Restate Cloud offers fully managed serverless with a free tier as the alternative. Both ends of the spectrum are genuinely available. Independent comparisons 2026-08-08 | Full |
| Prebuilt Agents, Templates & PacksShips SDKs and documented integration patterns including the Pydantic AI path, and IDEMPOTENCY IS ON BY DEFAULT which is a meaningful piece of prebuilt correctness most platforms leave to the developer, but there is no library of prebuilt agents or templates and none would be expected for a runtime. Independent comparison 2026-08-08 | Partial |
| Triggers & Channel CoverageWork is invoked through durable RPC and workflow entry points, sessions are keyed by user or conversation, and executions can pause and resume across arbitrary time spans - so the trigger surface covers both immediate invocation and long delayed resumption. Pydantic documentation 2026-08-08 | Full |
| Model Flexibility & RoutingRestate runs no models and makes no model selection; it provides durability for agents built elsewhere on models chosen elsewhere. Architecturally expected rather than a gap. Independent comparison 2026-08-08 | Unable to verify |
| APIs, SDKs & MCP ExtensibilityOpen source and code first with SDKs, a UI and an API for task control, and documented extension into third party agent frameworks through the Pydantic AI integration. No MCP support was retrieved. Pydantic documentation and independent comparison 2026-08-08 | Full |
| Testing, Debugging & OptimizationNo agent evaluation, testing or simulation capability is provided; evaluations are explicitly delegated to Logfire in the documented Pydantic pairing. The journal supports replay for recovery rather than for testing agent behaviour. Honest absence. Pydantic documentation 2026-08-08 | Unable to verify |
| Browser & Computer UseRestate executes code and coordinates services; no browser control, page navigation or computer use capability exists or would be expected of a durable runtime. Independent comparison 2026-08-08 | Unable to verify |
The Agentic Index coverage score grades every vendor Full, Partial or Unable to verify 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
Pricing
Free and open source self hosted; Restate Cloud has a free tier plus four paid plans
free open source self hosted; managed cloud plans not retrieved
Included quota
Unlimited when self hosted. Restate Cloud free tier limits were not retrieved.
What is public
The open source availability and the existence of a Restate Cloud free tier plus four paid plans are public. Specific rates were not retrieved in this pass.
Billing mechanics
Open source self hosted at no licence cost. Restate Cloud is consumption based serverless with a free tier and four paid plans, rates not retrieved.
Cost watchouts
**THE REAL COST HERE IS ADOPTION, NOT LICENCE. Independent comparison is blunt: the team must adopt Restate's service and Virtual Object model, and that is a real shift if the agents are plain Python loops. Budget the rewrite, not the runtime.** Self hosting is unusually cheap operationally for this category - a single binary with no separate database or worker processes - but it is still infrastructure the team runs. **AND NOTE THE CATEGORY WIDE TRAP THIS VENDOR HELPS ILLUSTRATE: durable execution platforms that meter per step create a paradox where more granular recovery means more billable executions. Restate self hosted sidesteps that entirely, which is part of the commercial argument for it.**
Variable cost rationale
Graded low because the self hosted path is free, complete and operationally light, so a team can run unlimited agent volume at infrastructure cost with no per step or per execution meter at all. That directly avoids the durable execution pricing paradox visible elsewhere in this pocket, where finer grained recovery produces more billable step executions. The unquantified item is Restate Cloud pricing, which buys managed convenience rather than access.
Additional watchouts
Model the migration effort to the service and Virtual Object model, which independent reviewers flag as the genuine cost. And note idempotency is on by default but keys do not deduplicate across separate requests.
Overage / add-ons
Not retrieved.
Sales call required
No, self serve available
Free / trial
Open source self hosted at no cost, plus a free tier on Restate Cloud
Lowest paid plan
Restate Cloud paid tiers, rates not retrieved
Commercial notes
**THE OPEN SOURCE POSITION IS UNUSUALLY STRONG BECAUSE THE SELF HOSTED PATH IS NOT A CRIPPLED VERSION: single binary, no extra databases, no separate worker processes, described independently as the least operationally demanding self hosted option in the durable execution category. That is a different open core shape from vendors who paywall governance, memory, model quality or the agents themselves - here the operational burden, not the capability, is what the managed tier removes.**
Key ambiguities
All Restate Cloud rates and the four plan structure. Given the self hosted path is free and genuinely complete, the cloud pricing determines only the managed convenience premium rather than access to the product.
Missing data
All Restate Cloud rates, free tier limits and the four plan structure.
Related vendors
- Acrab — Singapore compute infrastructure company building a full stack…
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- AIsa — Unified resource and payment gateway for AI agents that lets them…
- AlphaBitCore — AI control plane that governs how models, agents, tools, and…
- Anchor Browser — Cloud hosted browser infrastructure that lets AI agents operate real…
Alternatives to Restate
The closest documented capability profiles to Restate 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.
- Hatchet9.0 / 14Adds documented Testing, Debugging & Optimization
- Temporal9.5 / 14Adds documented Testing, Debugging & Optimization
- Pipedream9.0 / 14Adds documented Testing, Debugging & Optimization
- Windmill10.0 / 14Adds documented Model Flexibility & Routing and Testing, Debugging & Optimization
- BAND6.5 / 14Fuller documented coverage on Security, Identity & Governance
- Inngest10.5 / 14Adds documented Model Flexibility & Routing and Testing, Debugging & Optimization
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded