| Type | Startup | Startup | Startup |
|---|
| Categories | | | |
|---|
| In one line | Durable execution for long-running agent workflows. | The durable runtime for agents and backends. | Durable functions and workflows for AI agents. |
|---|
| Stage & fundingDiffers | $430M total · $300M Series D (Feb 2026) at $5B valuation | $27M total · $20M Series A (Sep 2026) | — |
|---|
| GitHub stars | — | — | — |
|---|
| License | — | — | — |
|---|
| Language | — | — | — |
|---|
| Last push | — | — | — |
|---|
| Works with | — | — | — |
|---|
| Built on | — | — | — |
|---|
| Built on it | — | — | — |
|---|
| Similar toDiffers | | | |
|---|
| People linkedDiffers | 2 | 3 | 2 |
|---|
| Events | 0 | 0 | 0 |
|---|
| Pros | - +MetadataMature, MIT-licensed durable-execution engine descended from Uber's Cadence, with about 24K GitHub stars and active development.[1][2]
- +DocsEvent history lets a crashed agent resume from its last recorded decision instead of starting over.[1]
- +User reportsPractitioners use it as the main building block for multi-step agents and value its real-time view of running workflows.[1][2]
| - +Independent testShips as a single binary with its own storage and replication layer, so a simple setup needs no external database or cluster.[1][2]
- +User reportsDurably commits each individual step before the next runs, which suits agent tool calls with side effects.[1]
- +DocsDesigned serverless-first, so handlers run on Lambda-style platforms without long-polling workers.[1]
- +User reportsUsers building agents report the TypeScript SDK is easy to use, and some pair it with Modal sandboxes for crash recovery.[1][2]
| - +User reportsOne team chose it over Temporal because it is much simpler to operate.[1]
- +User reportsDurable at the step level: builders say failed steps retry from that step rather than from scratch, which helps multi-step LLM pipelines that hit rate limits.[1][2]
- +DocsSelf-hosting has been supported since 1.0: a single server that defaults to SQLite and in-memory Redis, and can scale horizontally with external Postgres and Redis.[1]
- +User reportsA competitor's comparison recommends it for agents deployed on serverless platforms such as Vercel.[1]
|
|---|
| Cons | - −Independent testThere is a real learning curve around determinism and where to draw the workflow/activity line, and existing agent loops must be restructured to fit.[1]
- −DocsSelf-hosting means operating several Temporal cluster services, a persistence database and your own workers, and the main hidden cost is engineering time.[1][2]
- −DocsTemporal Cloud bills per action with a monthly minimum, which gets expensive for chatty, LLM-heavy agents.[1][2]
- −Independent testLarge LLM prompts and responses can saturate workflow history, which has size limits, forcing payload codecs that offload to external storage.[1]
| - −MetadataThe server is under the Business Source License 1.1, which forbids offering it as a public Restate platform service, so it is not OSI open source.[1]
- −MetadataMuch smaller ecosystem than Temporal: about 3K weekly npm downloads against about 25K, and about 4.6K GitHub stars.[1][2]
- −MetadataMost Restate-versus-Temporal comparisons come from Restate's own team or co-founders.[1][2]
- −User reportsA DBOS co-founder argues that running durable execution as a separate service is heavier to integrate than a library on Postgres.[1]
| - −MetadataThe server is licensed under the SSPL, converting to Apache-2.0 three years after release, so it is source-available rather than OSI open source.[1]
- −DocsEvery step counts as an execution (a run with 5 steps uses 6), and the free tier allows 50k executions, 5 concurrent steps and 24 hours of trace history.[1]
- −DocsThe single-node 'inngest start' mode is marked [Beta], and Inngest does not guarantee support for self-hosted instances.[1]
- −User reportsAt least one user called it 'unfortunately not very reliable'.[1]
- −User reportsThe cloud product sends your workflow data to Inngest, a point a competitor raised.[1]
|
|---|
| Best for | Teams running long-lived, crash-safe agent workflows that already have platform engineers to operate them. | Teams that want durable agent steps and state with a lighter footprint than a Temporal cluster, especially on serverless. | TypeScript and serverless teams that want durable, step-based agent workflows without running Temporal. |
|---|
| Watch out for | The determinism model and history-size limits when moving an existing agent loop onto Temporal. | BSL licensing and a smaller ecosystem. | Count steps, not runs, when you price it, and read the SSPL before self-hosting commercially. |
|---|
| Links | | | |
|---|