Alberto Miorin

Creator of BigConfig, the first agentic package manager.

[Alberto Miorin @amiorin]

The most interesting thing about `npx skills` isn't `npx`.

It's the word **skills**.

Skills package procedural knowledge — how to deploy, configure, validate, tear down — and make it reusable for an AI agent. That's the right direction.

But a skill only runs inside the agent.

The moment you need the same procedure in a terminal, in CI, in a cron job, or in a container — it's gone.

That's the gap BigConfig closes.

A BigConfig package is a Skill, expressed as software:

→ Shell-native, not agent-trapped. The same package runs from your terminal, your agent, your CI, your cron.

→ Software-grade reuse. Real dependencies, versioning, composition, testing.

→ A lifecycle contract — validate / build / create / describe / delete — so agents can safely describe and validate while humans stay in the loop for create and delete.

→ Durable artifacts. Terraform plans, lock files, validation reports. Output that survives the conversation and can be reviewed, diffed, and committed.

Procedural knowledge shouldn't only be written down for an agent. It should be executable, composable, testable, versioned — and usable from the shell.

BigConfig is the Open Agent Packages Ecosystem.

Like Skills, but executable.

👉 https://t.co/O9Ykdk0rCN

#AgenticDevOps #AI #DevOps #InfrastructureAsCode

[Alberto Miorin @amiorin]

DevOps is stuck in plan/apply/destroy.

Platform teams can keep Terraform, Helm, Ansible, and operators inside the package.

Delivery teams — and agents — should get an infrastructure program with verbs like validate, describe, backup, rollback, upgrade.

https://t.co/ymbPcdc3LB

[Alberto Miorin @amiorin]

New: the BigConfig Package Spec.

A small repo contract that `bc-pkg` can pin, initialize, and run: one manifest, one root `run` file, installable code/resources, and lifecycle verbs for humans, agents, CI, and shells.

Like Skills, but executable.

https://t.co/yrOJR12RVU

[Alberto Miorin @amiorin]

BigConfig now supports Clojure, TypeScript, and Python.

When we started, we built BigConfig in Clojure because we believed config-as-code deserved a real language. The challenge was always: how do you support more languages without splitting the ecosystem?

Coding agents changed the answer.

Agents can now maintain three first-class implementations in parallel — and the rendered config files themselves act as the test suite. If the output matches, the implementation is correct. No mocks, no parity bugs.

What this unlocks:
• Author packages in the language your team already uses
• One shared definition powers every implementation
• Infrastructure-as-code without the weight of CDK or Pulumi

Define once. Run everywhere.

👉 https://t.co/En0zAfvlNs

[Alberto Miorin @amiorin]

I recorded a short talk on **Agentic DevOps**.

The main idea:

> The agent advises. The operator acts.

Instead of giving AI agents production credentials, we can design infrastructure workflows around a safer four-verb contract:

- `validate` — read-only, safe for agents
- `describe` — read-only, safe for agents
- `create` — state-changing, operator-controlled
- `delete` — state-changing, operator-controlled

This gives agents a useful loop: inspect, validate, compare, and recommend — while humans keep control of production changes.

Slides: https://t.co/jWMmcEfNAY

Video: https://t.co/RYuOCmuqbX

#DevOps #AI #AgenticDevOps #Infrastructure #BigConfig

[Alberto Miorin @amiorin]

You should check out https://t.co/ZOR4bhBUVd

"Any cloud" means you can host 4 websites across 4 different cloud providers seamlessly. Your agent handles the infrastructure, not you—and it's completely safe.

Give it a spin:
`npx bc-pkg bigconfig-ai/once validate`

[Alberto Miorin @amiorin]

The agent has a production environment now.

For two years we've watched AI write code that never reaches a server. Demos. Sandboxes. "Works on my machine." The last mile — where software meets users — has been a manual checklist of YAML, DNS records, SMTP configs, firewall rules, and TLS certs that no agent could cross safely.

That ends today.

▸ Basecamp's ONCE showed us the way: software you install once, host yourself, and own forever. A one-line installer, a TUI dashboard, automatic updates and backups. No SaaS rent. No vendor lock-in. Your data on your server. What ONCE deliberately leaves out is the host itself — the cloud account, the DNS, the SMTP, the firewall.

▸ BigConfig is the first agentic package manager — a Clojure/Babashka framework for delivering infrastructure and software as a single, cohesive unit. Its most radical design choice is what it lacks: a dedicated CLI. Every package author defines their own task surface in bb.edn — the Clojure counterpart of package.json scripts. The agent writes the file. The agent runs the tasks. You stay in plain English.

▸ The Once BigConfig Package fills the gap ONCE leaves open. It turns any ONCE application into a single-command deployment: provisions a server (Hetzner Cloud, Oracle Cloud, or bring your own), wires up Cloudflare DNS, configures Resend SMTP end-to-end with DKIM and SPF, and runs Ansible playbooks to install Docker, firewall, and TLS. Two commands. bb once create. bb once delete. Reversible. No orphaned resources.

A vibecoder can now ship a real product without leaving Claude Code. Describe the app. Hand the agent a Docker image and a domain. The agent writes the bb.edn, provisions the box, points the domain, gets the certs, and hands you a working URL. Every step deterministic. Every step reproducible. Every step safe — because the agent isn't improvising shell, it's invoking packages with known contracts.

And it's turtles all the way down: BigConfig Marketplace is itself a ONCE application, deployed by the same once package it catalogs.

▸ Watch the demo: https://t.co/IbDWvzy4MX
▸ Browse packages and applications: https://t.co/M4VE8Td9iP
▸ Self-host your first app this weekend.

What would you ship if deployment were one command? Tell me below.

#AI #SelfHosting #ClaudeCode #Babashka #Clojure #DevOps #ONCE #BigConfig

[Alberto Miorin @amiorin]

Welcome to Once BigConfig Package Discussions!
https://t.co/LvZKyfXLdt

[Alberto Miorin @amiorin]

Welcome to BigConfig Discussions!
https://t.co/pSwCccTSNC

[Alberto Miorin @amiorin]

Revolutionizing DevOps: Meet BigConfig, the First Agentic Package Manager 🚀
I am excited to share a demo of BigConfig, a tool designed to re-imagine package management and software configuration for the era of AI agents.
In this walkthrough, you can see how an AI agent (Claude) takes the wheel to provision and deploy a full-stack environment with minimal human intervention.

Why BigConfig is a Game Changer:
▹ Agentic Automation: You don't need to learn the complexities of every tool; your agent operates BigConfig for you.
▹ Unified Orchestration: It seamlessly coordinates OpenTofu for infrastructure, Ansible for host configuration, and Docker/ONCE for application deployment.
▹ Clojure-Native & Powerfully Simple: Leverages the power of Clojure and Babashka to define complex workflows in a clean, scannable format.

The Demo in Action:
▹ The Setup: Claude is tasked with installing a website using a specific Docker image and domain.
▹ The Intelligence: The agent identifies missing parameters, asks clarifying questions about cloud providers (OCI), and handles secret management.
▹ The Execution: With a single bb once create command, BigConfig provisions an OCI VM, sets up Cloudflare DNS, configures Resend SMTP, and deploys the container.

The Result: A live, secured site at https://t.co/wZWfsLL6Tv.

The future of DevOps isn't just about better tools—it's about tools built to be driven by agents. 🤖✨

Check out the project on: https://t.co/tzvA7vGzjQ

#DevOps #AI #CloudComputing #BigConfig #AgenticWorkflow #Clojure #InfrastructureAsCode

[Alberto Miorin @amiorin]

@pesterhazy similar to https://t.co/qFiRzLImP2 . Finally, "dev and CI parity" is getting better.

[John Arundel @bitfield]

Reasons you might be interested in Bass:

you'd like to have a uniform stack between dev and CI
you're sick of YAML and want to write code instead of config
you'd like be able to audit or rebuild published artifacts
you're nostalgic about Lisp

https://t.co/F2sbcbBU7a

[Alberto Miorin @amiorin]

Open-Source Spotlight - OpenMetadata - Alberto Miorin https://t.co/7Z66egkR92 via @YouTube

[OpenMetadata @open_metadata]

If you are interested learning more about OpenMetadata and how to setup metadata ingestion, profiling & data quality tests please RSVP our webinar https://t.co/ucuHNnbFTY on March 23rd 9am PST

[OpenMetadata @open_metadata]

💖 we are sponsoring @samuelcolvin because of their fantastic work in building Pydantic. We in the @open_metadata community use Pydantic extensively, and along with JsonSchema, it made our ingestion framework pluggable and maintainable https://t.co/Jldtn0rDd7

[OpenMetadata @open_metadata]

OpenMetadata 0.9 release out!!
1. Activity Threads, Conversations to get data users to collaborate on data in a single place
2. Data Quality - Define and Run tests on data profiler metrics
3. Glossary - Controlled vocabulary to organize data.
https://t.co/IOubbP6kBB

[OpenMetadata @open_metadata]

Building Access Control policies in OpenMetadata https://t.co/QXNX77qcOS . Take a look at our design, road map behind the access control policies.

[Alberto Miorin @amiorin]

Please let your friends and colleagues know about OpenMetadata and also if you haven't starred our project on Github, please do so.

[Alberto Miorin @amiorin]

Our last meeting recording https://t.co/CTO2i1ytkN where we discuss the features and design for the next release.

Please add your thoughts in the following Github issues
https://t.co/APitdIWjwj
https://t.co/xzoe4Nkz13
https://t.co/TQjL2QBppY

[Alberto Miorin @amiorin]

@open_metadata is again on the Github Trending page. Thx everybody for your support and stars. Please continue to spread the word so that our project can reach an even larger audience. https://t.co/CYKIJEPgym