Vital signs
Why it's on the table
On the table, Orchestration (Or) is seat 13 of 58, in the Agents & Automation family. It is an experimental element — promising, volatile, and worth a contained experiment rather than a commitment. Budget curiosity, not dependence. It is optional: plenty of companies run without it — until a specific trigger (scale, regulation, cost, or customers) makes it essential for them. It sits in the lowest paid band — lunch money against the hours it returns.
Orchestration: the top 5 — v2026.Q3
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
1TemporalTemporal Technologies
Self-host free (MIT) · Cloud: Essentials from $100/mo (1M actions) · Business from $500/mo · Enterprise custom · $1,000 trial creditsBest for Production agent fleets and money-touching workflows that must survive crashes, redeploys, and week-long human-in-the-loop pauses.
The category owner by every measure: $300M Series D at $5B (a16z, Feb 17, 2026) on 380% YoY revenue growth, 9.1T lifetime Temporal Cloud actions — 1.86T of them from AI-native companies — and OpenAI in production calling durable execution 'a core requirement for modern AI systems.' The OpenAI Agents SDK integration went GA Mar 23, 2026, making crash-proof agents a decorator away.
Watch The steepest learning curve in the category — deterministic-workflow constraints, workers, task queues — and self-hosting the server is a real ops job. Per-action Cloud pricing gets chatty-agent expensive; every tool call and retry is an action.
2InngestInngest
Hobby free (50k executions/mo) · Pro from $99/mo (1M executions, 100 concurrency) · Enterprise customBest for TypeScript/serverless product teams who want durable step functions, events, and per-tenant flow control (debounce, throttle, concurrency) with zero infrastructure.
The developer-experience play: durable functions that run on your existing serverless or servers, triggered by events, with the best flow-control primitives in the category — the exact features multi-tenant AI products need. $21M Series A (Altimeter + a16z, Sep 16, 2025), SoundCloud and Day.ai in production, and 2026 shipped an agent stack — AgentKit for multi-agent networks plus Agent Evals, Sessions, and run scoring (Jun 2026).
Watch The server is SSPL with delayed Apache-2.0 — practically, you're buying the cloud. Execution-count pricing punishes chatty agents, and the eval push moves it onto Ev-element turf it hasn't proven yet.
3Trigger.devTrigger.dev (open source, Apache-2.0)
Free ($5/mo credit, 100k runs-ish) · Hobby $10/mo · Pro $50/mo + per-second compute (from $0.0000169/s) · Enterprise customBest for TypeScript agents that need long-running managed compute — no timeouts, checkpointed runs lasting hours — plus realtime streaming to your UI.
The only top pick that runs your code for you: tasks execute on Trigger.dev's elastic infrastructure with durable checkpointing, so agent runs survive crashes and last hours without you owning a worker fleet. 14.9k GitHub stars under Apache-2.0, v4 shipping steadily (v4.4.6, May 2026), and the whole product is now framed as 'build and deploy fully-managed AI agents.'
Watch TypeScript-only in practice, and the value is the managed compute — self-hosting the full platform is much heavier than the license suggests. Per-second machine pricing needs watching on long agent loops; smallest war chest of the top three.
4HatchetHatchet (YC W24)
MIT self-host free · Cloud: first 100k runs free then $10/1M · Team $500/mo + usage · Scale $1,000/mo · Enterprise customBest for Teams already on Postgres who want a fully open-source (MIT) task queue, DAG, and durable-workflow engine without adopting a new database or a restrictive license.
The clean open-source option: an MIT-licensed 'orchestration engine for background tasks, AI agents, and durable workflows' built on plain Postgres, with Go, Python, and TypeScript SDKs — 7.3k GitHub stars and steady releases (Jun 2026). Cheapest serious cloud entry in the category at $10 per million runs, and the Postgres-native thesis just got endorsed by Microsoft's pg_durable.
Watch The cloud jumps from hobby-cheap to $500/mo Team with 500 RPS and 3-day retention — mid-size pricing is lumpy. Small team, no disclosed funding beyond YC-stage, and Postgres becomes your scaling ceiling.
5RestateRestate (Stephan Ewen, Flink co-creator)
Single-binary self-host (BSL 1.1 → Apache after 4 yrs) · Cloud free 50k actions/mo · usage plans from $75/mo · BYOC capacity pricingBest for Latency-sensitive, stateful agents and transactional workflows — virtual objects and durable promises in one Rust binary instead of a cluster.
The deepest engineering of the challengers: a log-based Rust runtime (v1.6.2, Feb 2026) that fuses durable execution, state, and messaging, from the team behind Apache Flink — already running 'the account and credit card stack of a tier 1 bank.' Restate Cloud opened to everyone Sep 30, 2025, and Pydantic AI ships an official Restate durable-agents integration.
Watch Smallest community of the five (~4k stars) and BSL 1.1 is not open source — the license blocks offering Restate as a service. The architecture is novel enough that hiring for it is hard; funding beyond seed unverified.
Orchestration: the top 8 compared
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
| Tool | Execution model | License | SDK languages | Cloud entry | Self-host | Agent tooling | Proof at scale |
|---|---|---|---|---|---|---|---|
| Temporal | Server + workers (you run code) | MIT server | Go · Java · TS · Py · .NET · PHP · Ruby | $100/mo (1M actions) | Yes (real ops job) | OpenAI Agents SDK GA · AgentKit-style samples | OpenAI, Block; 9.1T actions |
| Inngest | Event-driven steps on your compute | SSPL server, Apache SDKs | TS · Py · Go · Kotlin/Java | $99/mo (1M execs) | Dev server yes; prod = cloud | AgentKit · Agent Evals · flow control | SoundCloud; Series A stage |
| Trigger.dev | Managed compute, checkpointed runs | Apache-2.0 | TypeScript | $10–50/mo + compute | Possible, heavy | AI SDK support · realtime streams | 14.9k stars; startup-heavy |
| Hatchet | Postgres engine + workers | MIT | Go · Py · TS (Ruby WIP) | 100k runs free, $10/1M | Yes (Postgres only) | Durable tasks for agents; DAGs | 7.3k stars; YC-stage |
| Restate | Log-based runtime, single binary | BSL 1.1 | TS · Java/Kotlin · Py · Go · Rust | free 50k · $75/mo | Yes (one binary) | Durable agents · Pydantic AI integration | Tier-1 bank card stack |
| DBOS | Library in your app + Postgres | MIT library; Conductor proprietary | Py · TS · Go · Java/Kotlin | Pro $99/mo (1M checkpoints) | Yes (lib is free) | OpenAI Agents SDK + Pydantic AI recovery | Smallest footprint approach |
| Orkes Conductor | Hosted state machines (Netflix Conductor) | OSS core (fork), enterprise cloud | Any (HTTP workers) · JSON DSL | custom (sales) | Yes (Conductor OSS) | LangGraph/CrewAI/ADK integrations | Claims 1B+ workflows/day |
| Vercel Workflows | 'use workflow' on Vercel Functions | OSS SDK, closed platform | TS/JS · Python | usage-based (events + data) | No | Agent guides; streams | New in 2026; multi-region |
How to choose your orchestration
- If agents touch money, customers, or anything you'd page someone over
- Temporal — the learning curve is the price of the only orchestrator with OpenAI-scale production proof. Budget real onboarding time and watch per-action costs on chatty agents.
- If you're a TS/serverless product team and your pain is events, retries, and per-tenant fairness (rate limits, debounce, concurrency)
- Inngest — its flow-control primitives are the category's best, and the free 50k-execution tier covers most pre-launch products.
- If your agent runs are long-lived compute (scraping, video, deep research) and you don't want to own workers
- Trigger.dev — the only top pick that sells you the machines too, with checkpointed no-timeout runs.
- If you already run Postgres and refuse non-OSS licenses
- Hatchet for a standalone MIT engine, or DBOS if you'd rather add a library to your app than a service to your stack — Microsoft's pg_durable (Jun 2026) says this whole approach is directionally right.
- If you're pre-PMF and tempted by all of the above
- Don't over-buy. A cron, a queue, and your deploy platform's built-in workflows (Vercel, Cloudflare) cover more than you think — adopt real orchestration when a failed retry first costs you money.
Orchestration: the whole field
24 more tools tracked in this category, including 4 dead, renamed, or sunsetting — a reference that hides the graveyard isn't one. Verified 2026-08-06.
| Tool | Maker | What it is | Entry | Status |
|---|---|---|---|---|
| DBOS | DBOS Inc. | Durable execution as a library — MIT Transact for Py/TS/Go/Java + your Postgres; OpenAI Agents SDK and Pydantic AI recovery integrations (Mar 2026); the near-miss for the top 5 | free lib · Pro $99/mo | active |
| Orkes Conductor | Orkes | Enterprise cloud on Netflix Conductor; agentic workflows spanning LangGraph/CrewAI/OpenAI/ADK; claims 1B+ workflows/day (vendor) | free trial · sales | active |
| Conductor OSS | Netflix → community/Orkes | Netflix archived its repo Dec 2023; development continued in the community conductor-oss fork stewarded by Orkes | free | renamed |
| Cadence | Uber | Temporal's direct ancestor, still maintained and used inside Uber; choose Temporal unless you're Uber | free | active |
| AWS Step Functions | AWS | Managed state machines, JSON-defined; rock-solid but agent ergonomics trail code-first rivals; pairs with Bedrock AgentCore | per state transition | active |
| Azure Durable Functions / Durable Task Scheduler | Microsoft | Code-first durable orchestration in .NET/Python/JS; Durable Task Scheduler adds managed backend; feeds Agent Framework durable workflows | consumption pricing | active |
| pg_durable | Microsoft | Open-sourced Jun 2026: durable execution inside Postgres itself (Rust duroxide libs) — preview, but a loud endorsement of the DBOS/Hatchet thesis | free (preview) | active |
| Google Cloud Workflows | YAML-defined serverless orchestration; fine glue for GCP APIs, not an agent runtime | per step, free tier | active | |
| Cloudflare Workflows | Cloudflare | Durable multi-step execution on Workers, free-plan included; underpins Cloudflare's Agents SDK — the cheapest on-ramp to durability | free + usage | active |
| Vercel Workflows | Vercel | 'use workflow' directive on the open Workflow SDK; TS + Python, multi-region, managed persistence (2026) — orchestration bundled into the deploy platform | usage-based | active |
| Upstash Workflow | Upstash | QStash-based durable steps for serverless TS/Py; pay-per-message, generous free tier | free tier + usage | active |
| Windmill | Windmill Labs (YC S22) | Open-source code-first scripts/flows/apps platform, 4,000+ orgs; straddles this element and Wf — pick it when humans and jobs share one tool | free · cloud + EE | active |
| Kestra | Kestra | Declarative YAML event-driven orchestration, strong data-eng adoption; agent features arriving but data pipelines remain the center | free · EE sales | active |
| Apache Airflow | Apache / Astronomer | The data-pipeline scheduler standard (3.x); wrong shape for agent loops — batch DAGs, not durable long-running state | free · Astro cloud | active |
| Prefect | Prefect | Pythonic dataflow orchestration; ControlFlow agent experiments folded back toward core data workloads | free tier | active |
| Dagster | Dagster Labs | Asset-oriented data orchestration; excellent for pipelines feeding agents (embeddings, ETL), not for running them | free tier | active |
| Celery | community | The Python task-queue incumbent; no durable state, no replay — the thing this category replaces | free | active |
| BullMQ | Taskforce.sh | Redis-backed Node job queue; ubiquitous, but durability ends at the queue | free · Pro paid | active |
| River | River (Brandur Leach et al.) | Fast Postgres-native Go job queue with a workflow layer growing on top | free · Pro paid | active |
| Resonate | Resonate HQ | Durable promises as an open spec/runtime — elegant model, small adoption | free | active |
| Hyrex | Hyrex (YC) | Postgres-backed Python/TS task orchestration entrant in the Hatchet lane | free + cloud | active |
| Defer | Defer (France) | TypeScript background-jobs platform; shut down May 2024 — the category's first casualty, users migrated to Trigger.dev/Inngest | — | dead |
| Mergent | Mergent → Resend | Task/cron API acquired by Resend (2024); product absorbed | — | acquired |
| Quirrel | Simon Knott → Netlify | OSS job queue for serverless; creator acqui-hired by Netlify (2022), project effectively unmaintained | — | dead |
Orchestration: the category in numbers
Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.
- Temporal raised $300M Series D at $5B (a16z, Feb 17, 2026) — the category's first mega-round — on 380% YoY revenue growth and 20M+ monthly installs; OpenAI: 'durability is as important as performance' [src]
- AI is now the growth engine of orchestration: 1.86T of Temporal Cloud's 9.1T lifetime actions came from AI-native companies (Feb 2026); Temporal had raised $146M at just $1.72B eleven months earlier (Mar 2025) — a ~3x valuation jump in under a year [src]
- Deploy platforms are absorbing entry-level orchestration: Vercel Workflows ('use workflow', TS + Python, multi-region, 2026) and Cloudflare Workflows (GA, on the free plan) bundle durable execution into hosting [src]
- The Postgres-native wave crested: Microsoft open-sourced pg_durable (Jun 2026), putting durable execution inside Postgres — validating the DBOS/Hatchet/River approach and commoditizing the low end [src]
- Inngest raised a $21M Series A (Altimeter + a16z, Sep 16, 2025) and by Jun 2026 was shipping agent evals and run-scoring — orchestrators are moving up-stack into observability/evals territory [src]
- Every serious agent framework now ships durable-execution integrations: OpenAI Agents SDK (Temporal GA Mar 2026, DBOS), Pydantic AI (Temporal, Restate, DBOS), LangGraph (its own checkpointer) — durability became the assumed substrate, not a feature [src]
Orchestration: method & sources
Scope: durable execution and job orchestration for agentic workloads. No-code/visual automation (n8n, Zapier, Make) is element Wf; in-process agent graphs and checkpointers (LangGraph, CrewAI, OpenAI Agents SDK itself) are element Ag — this element is the layer that keeps those graphs alive across crashes and weeks. Inngest's Agent Evals overlaps element Ev. Data-pipeline orchestrators (Airflow, Prefect, Dagster, Kestra) are listed in the field for disambiguation but are a different job. Conflicts resolved: Temporal's $1.72B (Mar 2025) and $5B (Feb 2026) valuations are two successive rounds, not a discrepancy — we cite the later. The pg_durable story is sourced from an aggregator (agent-wars.com) plus the GitHub repo; the repo URL and 'preview' status should be re-checked. Orkes's '1B+ workflows/day' and Restate's 'BYOC up to 10x cheaper' are vendor claims, unaudited. Restate post-seed funding could not be verified from primary sources and is omitted. Inngest's own blog index mislabels its Series A as Sep 2026; the post itself is dated Sep 16, 2025 — we use the post. Canon maturity is 'Experimental' and the ranking criteria were: production proof at scale > agent-specific tooling > licensing/self-host freedom > price of entry. Ranking criteria: verified commercial traction, independent satisfaction surveys, agent benchmarks, and founder-fit (price floor, lock-in, surfaces). Editorial, never paid — the charter. Machine-readable twin: or.json.
All sources (24)
- https://temporal.io/blog/temporal-raises-usd300m-series-d-at-a-usd5b-valuation
- https://temporal.io/news/temporal-technologies-secures-valuation-to-fuel-durable-execution
- https://temporal.io/blog/announcing-openai-agents-sdk-integration
- https://temporal.io/pricing
- https://github.com/temporalio/temporal
- https://www.inngest.com/blog/announcing-inngest-series-a
- https://www.inngest.com/pricing
- https://github.com/inngest/inngest
- https://agentkit.inngest.com
- https://trigger.dev/pricing
- https://github.com/triggerdotdev/trigger.dev
- https://hatchet.run/pricing
- https://github.com/hatchet-dev/hatchet
- https://www.restate.dev/blog/announcing-restate-cloud-public
- https://restate.dev/cloud
- https://github.com/restatedev/restate
- https://pydantic.dev/articles/restate-durable-execution-pydanticai
- https://www.dbos.dev/pricing
- https://www.dbos.dev/blog/dbos-new-features-march-2026
- https://vercel.com/docs/workflows
- https://developers.cloudflare.com/workflows/
- https://orkes.io
- https://www.windmill.dev
- https://agent-wars.com/news/2026-06-07-microsoft-pg-durable
Our take
The element most likely to matter more next year than this one. Watch it; don't over-buy it yet.
Combines with
Appears in compounds
This is element 13 of 58. The table is versioned quarterly — when a tool loses its seat, the changelog records the succession.
Explore the full table →