autokitteh

Open-source workflow automation, honestly reviewed. No marketing fluff, just what you get when you self-host it.

TL;DR

• What it is: Open-source (Apache-2.0) durable workflow automation platform — write vanilla Python, autokitteh makes it survive server crashes, retries, and long-running steps without you managing state [README].
• Who it’s for: Backend engineers and DevOps teams building workflows that must not lose state — think multi-step approval flows, SOAR (security orchestration), MLOps pipelines, FinOps automations. Not for non-technical founders. [README][2].
• Cost savings: Zapier charges per task. autokitteh self-hosted is Apache-2.0 — run it on your own server, pay nothing per execution. Cloud pricing is not published; the offering was in beta as of this writing [README].
• Key strength: Durable execution backed by Temporal — if your server crashes mid-workflow, autokitteh resumes exactly where it stopped, with no lost state and no zombie processes [3][README].
• Key weakness: This is a code-first tool. There is no drag-and-drop builder. You write Python. If your team can’t read a for loop, this is the wrong product [README][2].

---

What is autokitteh

autokitteh is a workflow automation platform for developers. You write a Python function; autokitteh runs it as a durable, long-running workflow. The GitHub description is a good summary: “Durable workflow automation in just a few lines of code” [README].

Under the hood it’s built on Temporal, which is the same infrastructure layer that Stripe, Netflix, and Coinbase use for reliable distributed workflows. Temporal handles the hard parts — retrying failed steps, persisting workflow state, distributing execution across workers. autokitteh wraps Temporal with a friendlier developer experience: a CLI (ak), a VS Code extension, a web UI, pre-built integrations for Slack, GitHub, Gmail, Twilio, ChatGPT, Gemini, and others, and a Python/Starlark runtime [README][3][4].

The pitch it makes against Zapier, n8n, and Make.com is blunt: those tools are no-code/low-code drag-and-drop builders that hit a ceiling fast. autokitteh is for teams that already know they need code-level flexibility and are tired of writing boilerplate durability logic on top of something like raw Temporal or AWS Step Functions [README].

As of this review, autokitteh sits at 1,107 GitHub stars — small relative to n8n (100K+) or Activepieces (21K+), which tells you something about the audience size. This is not a mainstream tool yet. It’s a niche product aimed at a specific kind of engineering team.

---

Why people choose it

The articles that mention autokitteh don’t do deep feature reviews — they mention it alongside n8n and windmill.dev as one of three credible self-hosted platforms for handling AI automation with durable workflows [2].

The core argument [2] is about interruption cost and reliability: modern AI-driven workflows that route notifications, summarize email, and monitor system state can’t afford to lose their place if a container restarts. A naive implementation using a simple scheduler or cron job will either duplicate work or silently drop state on crashes. Temporal-based systems like autokitteh don’t have that problem — they write-ahead-log everything and resume from the exact step where execution stopped [3][README].

The comparison from the rogverse article [2] is instructive:

• n8n — visual automation, 400+ integrations, good for teams who want drag-and-drop. Best for visual workflow design.
• windmill.dev — script-to-production with auto-generated UIs and APIs, Python/TypeScript/Go. Best for developer teams who want internal tools.
• autokitteh — long-running, multi-step workflows with Temporal durability. Zero data loss on failures. Best when you need guaranteed execution, not just probably reliable execution [2].

The medevel.com roundup [1] slots autokitteh specifically in the event-driven category — systems that react to state changes and maintain context across time, not just fire-and-forget task runners. The quickstart demo [3] makes this concrete: a workflow that loops 50 iterations with a sleep between each step, survives a server kill mid-loop, and resumes from where it was. That’s the capability no visual no-code tool delivers without significant custom infrastructure.

What pulls developers toward it over raw Temporal: autokitteh hides the scheduling complexity, the gRPC boilerplate, the worker pool management. You write vanilla Python. The platform handles persistence [README][3].

---

Features

Core execution engine:

• Write workflows in Python or Starlark (a deterministic Python dialect); JavaScript support listed as “coming soon” [README]
• Durable execution: state is persisted across server restarts, network failures, and partial crashes [3][README]
• Automated recovery without state loss — the resilience demo in the quickstart [3] shows this concretely
• Long-running workflows: a workflow can wait for a human approval for days without holding a thread
• Schedulers and webhooks as triggers [README][3]
• gRPC/HTTP API — described as “API-first” with all services accessible programmatically [README]

Interfaces:

• CLI tool (ak) — install via Homebrew on macOS, build from source on other platforms [3][README]
• VS Code extension — build and manage workflows from the editor, includes LSP autocomplete for script code [4]
• Web UI for management, monitoring, and debugging [README]

Built-in integrations:

• Slack, GitHub, Twilio, ChatGPT (OpenAI), Gemini, Gmail, Google Calendar, HTTP, gRPC listed in README [README]
• Additional integrations available through the kittehub example repository
• New integrations can be added; the platform is designed for extensibility [README]

Deployment options:

• Self-hosted (open-source server, primary use case) [README]
• Managed cloud iPaaS (beta at time of writing; contact [email protected]) [README]
• Docker-based local dev mode via ak up --mode dev [3]

Use cases cited:

• DevOps and MLOps automation
• FinOps workflows
• SOAR (Security Orchestration, Automation, and Response)
• Productivity and notification management [README][2]

What’s missing:

• No drag-and-drop visual builder
• No large pre-built integration marketplace comparable to n8n or Zapier
• No MCP (Model Context Protocol) support mentioned anywhere in the available sources
• JavaScript runtime listed as “coming soon” — not available yet [README]

---

Pricing: SaaS vs self-hosted math

Self-hosted (open-source):

• Software: $0 (Apache-2.0 license) [README]
• Infrastructure: $5–20/month VPS depending on workflow volume
• Requires Go 1.24 to build from source, or Docker for the dev server [README][3]

autokitteh Cloud:

• Pricing not published. Cloud offering described as “currently in beta” in the README; contact [email protected] for details [README]
• A free trial exists at autokitteh.cloud but tier limits and costs are not available from any source reviewed

Zapier for comparison:

• Free: 5 Zaps, 100 tasks/month
• Starter: $19.99/mo for 750 tasks
• Professional: $49/mo for 2,000 tasks; scales past $100/mo at volume

n8n for comparison:

• Self-hosted: $0 (Fair-code Sustainable Use License, restricts commercial redistribution)
• n8n Cloud: starts around $20/mo for 2,500 executions/month

The honest pricing picture for autokitteh: if you self-host, it’s free to run. If you want cloud, you don’t know what it costs without contacting sales. For a non-technical founder evaluating tools, this opacity is a red flag. For an engineering team with a DevOps budget and the ability to self-host, it’s irrelevant — the Apache-2.0 license means you can run it in your own infrastructure indefinitely with no commercial restrictions.

---

Deployment reality check

The quickstart [3] is honest about what’s required. The path is:

1. Install ak CLI via Homebrew (brew install autokitteh/tap/autokitteh)
2. Run ak up --mode dev to start a local server
3. Clone kittehub, deploy a project with ak deploy --manifest ...
4. Trigger via webhook curl

For production, you’d need Docker or a Go 1.24 build environment, a domain with HTTPS (for OAuth flows and webhook reception), and either a managed database or the embedded one. The README lists build requirements including buf, docker, go >= 1.24, and golangci-lint for full builds — this is not a one-click Heroku deploy [README].

For third-party integrations like Slack and GitHub, autokitteh needs publicly accessible HTTPS endpoints for OAuth 2.0 and webhook delivery. The docs suggest ngrok or similar tunneling tools for local development [3]. On a cloud VPS this is standard nginx + Let’s Encrypt territory.

The resilience demo [3] is the best argument for the platform’s core value prop. You start a 50-iteration loop with 1-second sleeps, kill the server mid-run (pkill -f ak up), restart it, and the workflow resumes from where it stopped. This is the behavior most people have to build themselves on top of Redis or a database with a homemade checkpoint system. autokitteh gives it to you by default.

Realistic time estimate: A developer familiar with Python, Docker, and self-hosted tooling: 1–2 hours to a working local instance, half a day to production with TLS and an integration like Slack. No relevant path for non-technical founders — this requires coding.

---

Pros and cons

Pros

• Apache-2.0 license — genuinely permissive. Self-host, fork, embed in commercial products, no restrictions. Better than n8n’s Fair-code license for teams that want to build on top of it [README].
• Temporal-backed durability — workflows survive server crashes with zero state loss. This is a real engineering capability, not a marketing claim — the quickstart demo [3] is reproducible proof.
• Write real Python — no visual flow builder to fight, no proprietary DSL to learn. Workflows are code, versioned in git, reviewable in PRs, testable with standard tools [README].
• API-first — everything accessible via gRPC/HTTP. Automation platforms that lock you into a UI are limiting; autokitteh treats the API as the primary interface [README].
• VS Code extension with LSP autocomplete — lowers the friction for developers building integrations [4].
• Kittehub example repository — a practical library of working automation examples to start from rather than a blank canvas.
• No per-execution pricing on self-hosted — workflows can run millions of times, durably, for the cost of your VPS.

Cons

• Code-only — there is no drag-and-drop builder. This is a deliberate design choice, not an oversight, but it draws a hard line: if your team can’t write Python, they can’t use autokitteh [README].
• Small integration catalog — Slack, GitHub, Gmail, Twilio, a few AI providers. n8n has 400+ nodes; Zapier has thousands. For long-tail SaaS integrations you’re writing HTTP calls yourself [README].
• Cloud pricing opaque — “currently in beta, contact us” is not a pricing page. If you’re evaluating SaaS options, you can’t compare costs [README].
• Very low adoption — 1,107 GitHub stars as of this writing. By comparison n8n has 100K+. Smaller community means fewer Stack Overflow answers, fewer pre-built examples, more likelihood of hitting unresolved issues without a fast response.
• JavaScript listed as “coming soon” — if your team writes TypeScript and not Python, you’re waiting for a feature that hasn’t shipped [README].
• No MCP support — for teams building Claude Desktop integrations or AI agent pipelines that expect MCP-native tools, autokitteh isn’t there yet.
• No drag-and-drop visual debugging — Make.com and n8n both show you step-by-step what each node processed. autokitteh’s debugging is through session logs in the CLI or web UI, which is less accessible for non-engineers.
• No community reviews found — no Trustpilot, no G2, no Reddit threads with user accounts of pain points or production experience. You’re evaluating a product with almost no third-party validation.

---

Who should use this / who shouldn’t

Use autokitteh if:

• You’re a backend engineer or DevOps team who already writes Python and needs workflows that must not lose state on failure.
• You’re building SOAR automations, MLOps pipelines, or FinOps workflows where “it probably ran” is not acceptable — you need “it ran, exactly once, and here’s the audit log.”
• You want Apache-2.0 freedom to embed durable workflow infrastructure in a commercial product without license negotiations.
• Your team is comfortable running self-hosted infrastructure and doesn’t need a managed SaaS.
• You’ve hit the ceiling of cron jobs and simple webhooks and need proper workflow orchestration without the complexity of raw Temporal.

Skip it (pick n8n instead) if:

• You want a large integration catalog and a visual builder your whole team can use, not just engineers.
• You need a drag-and-drop interface for marketing, HR, or operations staff.
• You prefer a tool with a massive community and years of Stack Overflow answers.

Skip it (pick windmill.dev instead) if:

• You want code-first automation (Python, TypeScript, Go, SQL) but also need auto-generated UIs, internal tools, and a broader developer experience platform [2].

Skip it (stay on Zapier) if:

• Your team is non-technical and the visual builder is non-negotiable.
• You need instant access to 1,000+ pre-built integrations without writing any code.
• You don’t have anyone who can manage self-hosted infrastructure.

Skip it entirely for now if:

• You need to know what the cloud costs before you can evaluate — the pricing opacity may resolve as they exit beta, but it’s a blocker for budget-constrained evaluation.

---

Alternatives worth considering

• n8n — the realistic alternative for most teams. Fair-code license (more restrictive than Apache-2.0 for commercial embedding), visual builder, 400+ integrations, strong community. If you want to hand workflows to non-engineers, n8n is the right tool.
• windmill.dev — code-first like autokitteh but broader scope: auto-generated UIs, REST API generation, multi-language (Python, TypeScript, Go, SQL, Bash). Better choice if you want a full internal tools platform, not just workflow orchestration [2].
• Temporal directly — if you’re already building microservices with Go or Java and want durable execution at scale, run raw Temporal. autokitteh is Temporal with a developer experience layer; if you have the engineering bandwidth to run Temporal natively, you get more control [README].
• Zapier — the incumbent. No self-hosting, no code, easiest onboarding, most expensive at scale, fully closed source. Still the right answer for non-technical founders with simple integrations and short task lists.
• Prefect / Dagster — Python-native workflow orchestration platforms popular for data and ML pipelines. If your use case is primarily data engineering rather than general-purpose automation, these may be better fits.
• Activepieces — MIT-licensed, visual builder, solid Zapier replacement for non-technical teams. The right choice when your team doesn’t write code.

---

Bottom line

autokitteh solves a real problem: Python workflows that must keep running correctly even when infrastructure fails. Built on Temporal, it delivers production-grade durability that would otherwise require significant engineering investment to replicate. The Apache-2.0 license means you can build on it commercially without calling a lawyer.

But it is not for non-technical founders escaping SaaS bills. It’s for developers who write Python, need orchestration that survives crashes, and are comfortable running self-hosted infrastructure. The tiny integration catalog and small community are genuine constraints today. If those constraints are acceptable for your use case — SOAR, MLOps, multi-step approval flows, FinOps automation — autokitteh delivers something few tools in this space do: code you wrote, running reliably on infrastructure you own, resuming exactly where it stopped when things go wrong.

If you need a self-hosted Zapier replacement, look at Activepieces or n8n. If you need durable Python automation with no per-execution cost and no vendor lock-in, autokitteh is worth the afternoon to evaluate.

---

Sources

1. medevel.com — “20 Killer Open-Source Self-hosted Alternatives to n8n: Event-Driven vs Task-Driven Automation Unleashed!” https://medevel.com/20-killer-open-source-self-hosted-alternatives-to-n8n-event-driven-vs-task-driven-automation-unleashed/

2. roger.rogverse.fyi — “Stop Grabbing Your Phone Every Time It Beeps: Using AI Automation with Durable Workflows” (Dec 9, 2025). https://roger.rogverse.fyi/stop-grabbing-your-phone-every-time-it-beeps-why-using-ai-automation-withdurable-workflows-is-non-negotiable.html

3. docs.autokitteh.com — “Quickstart | AutoKitteh”. https://docs.autokitteh.com/get_started/quickstart

4. docs.autokitteh.com — “Installing the VS Code Extension | AutoKitteh”. https://docs.autokitteh.com/get_started/vscode/install

Primary sources:

• GitHub repository and README: https://github.com/autokitteh/autokitteh (1,107 stars, Apache-2.0 license)
• Official website: https://autokitteh.com
• Cloud offering: https://autokitteh.cloud
• Example automations: https://github.com/autokitteh/kittehub