A practical guide to the parts of senro you use every day: checking code out into a workspace, caching a step so it never runs twice, fanning one step out over every module in a monorepo, narrowing that fan-out to only what a change affects, and letting the pipeline binary decide for itself whether a webhook is its business.
CI systems ask you to describe real programming (dependency graphs, retries, branching, recovery) in YAML, a language with no functions, no types, no tests and no debugger. senro is a new open-source pipeline engine that starts from the other end: the pipeline is the Go program. It builds an immutable plan, runs it locally, in containers, on Kubernetes or over SSH, and exposes one event stream you can attach to live.
Debugging a CI failure usually means adding a print statement and pushing again. senro records every run as an append-only event stream you can replay with the same client that watched it live, lets you open a shell inside a step while it is still running, explains cache misses component by component, and will re-execute a step to check whether it lied about being pure. This is the field guide, in the order you should use it.
A practical comparison of the two main Go AI frameworks, built around the same production service written twice: RAG, tool calling, typed output, retries, cross-vendor fallback and tracing. 73 lines in Genkit Go against 272 in LangChainGo.
A complete, step-by-step migration guide. Every LangChainGo concept mapped to its Genkit Go equivalent, with before and after code for models, prompts, chains, structured output, tools, agents, RAG and memory.
A Genkit flow is a great unit of AI work and a poor unit of orchestration. senro is a pipeline engine where every step already has retries, caching, an event stream and live attach. Put a flow inside a step and each model call becomes a graph node: change one document out of three and you pay for one call, not three. A practical guide, with every example running against the real Gemini API.
A companion to my Genkit Go vs ADK Go 2.0 comparison, this time only about agents. I built a fire-and-forget research team in 90 lines of Genkit: an orchestrator with two sub-agents, artifacts, cross-vendor model fallback and human approval, then tried to build the same thing in ADK. Middleware and the plugin ecosystem are what make the difference, and every program in this article ran.
I built the same eight programs twice, once with Genkit Go and once with Google’s ADK Go 2.0, and ran them all against Gemini: hello world, structured output, streaming, tools, middleware, custom models, multi-turn and HTTP serving. This is what the code actually looks like in each, where ADK makes you pay an “agent tax”, where Genkit lets you drop lower, and where ADK’s ceremony genuinely pays off.
You updated a Kubernetes Secret and then did the ritual: kubectl rollout restart. There is a better way. mamori watches your Secrets and ConfigMaps with native informers and reconciles the change into your running Go process, atomically and validated, without bouncing a single pod.
In GCP Secret Manager, secret versions are immutable and you rotate by adding a new version and moving the “latest” alias. But your Go service resolved “latest” once at boot and froze it there. mamori watches the secret, notices when “latest” moves, and reconciles the new version into your running process, validated and redacted.
Azure Key Vault versions your secrets and lets you rotate them behind the scenes, but your Go service resolved the secret once at boot and never looked again. mamori watches Key Vault and reconciles a new version into your running process, typed, validated, and redacted, without a restart.
AWS Secrets Manager will happily rotate your database password every 30 days. The problem is your Go service, which read that password once at boot and has no idea it changed. mamori closes that gap: it watches Secrets Manager and Parameter Store and reconciles rotations into your running process, typed, validated, and redacted.
Every Go service ends up hand-rolling the same thing: a config loader, a secrets fetcher, a ticker to refresh them, a mutex, and a prayer that nothing leaks into the logs. mamori is a new open-source Go library that loads configuration and secrets from more than 30 sources into typed, validated structs, and keeps them reconciled at runtime, without a restart.
Genkit flows are great at orchestrating LLMs, tools and RAG, but they live and die with the process that runs them. The new genkitx-temporal plugin lets you run any Genkit flow as a Temporal Workflow, giving you retries, durable history, timeouts, cancellation and a UI to inspect every execution.
Building AI applications without a local debugging tool is like writing backend code without a debugger. A look at the “shift-left” philosophy applied to Gen AI development, and a hands-on comparison of Genkit Developer UI, Vercel AI SDK DevTools and Mastra Studio.
A side-by-side comparison of the two leading middleware systems in the JS/TS Gen AI ecosystem: Vercel AI SDK’s wrapLanguageModel and Genkit’s generateMiddleware. APIs, mental model, built-ins, composition, observability and when to pick each.
A deep dive into the new Genkit middleware system for JavaScript/TypeScript: built-in middleware (filesystem, skills, toolApproval, retry, fallback), how to build your own with generateMiddleware, and the new model/tool/generate interception hooks.
Aprende a empezar con Genkit en JS/TS desde cero: instalacion, primer flow, ejecucion local y uso de la Developer UI.
Python has dominated the Gen AI conversation, but it is not the only, nor the best, option for production. Here is why Go (and Genkit Go in particular) is a stronger bet for serious AI services in 2026.
A practical, in-depth comparison of the top Generative AI frameworks in 2026: Genkit, Vercel AI SDK, Mastra, LangChain, and Google ADK, from someone who has built with all of them. (English)
