Kestra

La team Kestra en direct du stand au DevLille ! 🤘🏻🎉

When Vi­ssimo's e-commerce integrations were scattered across lambdas, crons, and an iPaaS, a Black Friday incident meant log hunts across five systems and manual order queue reprocessing at 3am. Scattered integrations treat failures as exceptional. The fix was treating them as operational: event-driven triggers define the contract at the boundary, declared retries and timeouts replace improvised recovery, and idempotency by business key makes reprocessing repeatable. The integration layer dropped to ~10% of previous iPaaS spend. Black Friday 2025 ran without a critical incident. Rafael B., Engineering and Architecture Manager at Vi­ssimo Group, walks through the full migration in Kestra Engineering on Medium, where we publish technical learnings from engineers building with Kestra. https://lnkd.in/e_s_mvae

Alex Emerich is Kestra's entire documentation team. He covers 1,400+ plugin references, all product docs, how-to guides, and the release blog post every cycle. A community member recently called it the best documentation they'd ever come across. He keeps pace with a daily-shipping engineering team by offloading the research layer to AI: which PRs actually require docs, what source code says before prose goes in, whether existing pages still describe current behavior. Next Thursday, AJ and Will Russell are running a live Q&A on exactly that workflow. What the triage layer looks like, how source code research works in practice, and how reusable skills enforce standards across a surface too large to audit by hand. If you write docs or maintain them alongside code, you'll leave knowing which parts of the work belong to AI (triage, style checking, link validation) and which parts still require a writer. Register today: https://lnkd.in/eehzzEre

A community member recently called Kestra’s documentation the best they’d ever come across. One person manages all of it: 1,400+ plugin references, all product docs, and the release blog post every cycle. When your engineering team ships daily, the hardest part of documentation isn’t writing. It’s triage: figuring out which PRs need docs, which existing pages just went stale, and whether the reference you wrote last sprint still describes how the product actually behaves. Join AJ Emerich (Technical Writer) and Will Russell (Developer Advocate) at Kestra for a live Q&A on the workflow AJ built to stay ahead of it. You’ll learn the best practices AJ uses to triage PRs for docs impact, trace source code behavior before writing a word, and enforce editorial standards automatically across hundreds of pages. If you write docs or maintain them alongside code, you’ll leave knowing which parts of the work belong to AI (triage, style checking, link validation) and which parts still require a writer.

Most RAG pipeline work happens before the model runs: parsing documents, chunking text, generating embeddings, indexing vectors, validating quality. Stitching those layers together is where most engineering time goes. We mapped the Kestra plugins covering each step - from Apache Tika for document ingestion to Weaviate for vector storage. Every step is a YAML task, version-controlled and observable. François Delbrayelle covers the full ecosystem in his latest post here: https://lnkd.in/ehQfAckS

When a data engineer at a French bank kept email notification logic inside an inline Python task, a semicolon in the recipient list sent half the recipients dark, with executions showing success throughout. Inline Python handles edge cases fast and hides orchestration failures behind green executions. The fix was moving responsibility: Kestra owns secrets, inputs, outputs, retries, and logs. Python owns what only Python could do: build the payload, sign the request, send it, write the output. Forty lines of Python became a dozen lines of YAML and a focused adapter script. Daniel ALLANIC walks through the full incident in Kestra Engineering on Medium, where we publish technical patterns from engineers building with Kestra. Read here: https://lnkd.in/e_B5-JAy

The VM ticket might be infra's most pointless ritual. You wait 5 days for a VM provisioning, then watch it idle at 8% CPU for a year. None of it is real work; it's a workflow we keep pretending is a ticket. None of that needs a human. So I wrote up doing the whole thing in Kestra approval triggers it, one flow for any cloud, idle VMs get cleaned up automatically, and it's mostly just Pebble expressions {{ }}, barely any code. more here: https://lnkd.in/esws5_Bv

AI prototypes come together quickly. Getting one to production is a different problem: scheduling, retries, human-in-loop checkpoints before outputs get committed, and observability across the full pipeline. How do you actually build for all of that? We're partnering with DataTalksClub on LLM Zoomcamp 2026 to answer exactly that: a free 10-week course where you'll build RAG pipelines, AI agents, and multi-agent systems from scratch. Getting the agentic orchestration piece right is critical to shipping any of it reliably, which is why we built the module that covers it. Class starts today, sign up here: https://lnkd.in/dZVfkmvX

With the LLM Zoomcamp by DataTalksClub launching next week, it only felt appropriate to write about RAG and how you can actually put it into production. It can be straightforward to get working inside of a notebook, but it can feel like a big step to run it in production. In this blog post, I walk through how orchestration can help us with this, and allow us to build a full RAG pipeline that handles each part of the process! Link in the comments 👇

The Broadcom era broke a 15-year habit: nobody used to think about their hypervisor. Now everyone is. Perpetual licenses are gone, costs are up 3–5x, and Gartner expects 35% of VMware workloads to move off-platform by 2028. So teams are scrambling toward Proxmox, Nutanix, XCP-ng, OpenShift Virtualization… But here's the trap: swapping VMware + Aria for Nutanix + Prism is just new lock-in. My latest piece makes the case for an orchestration-first strategy, keep the hypervisor swappable, and let Kestra coordinate provisioning and migration across any platform. Change targets by editing one block, not rewriting every pipeline. Full deep-dive: https://lnkd.in/dvjpeCwP