Data pipelines
Ingestion, transformation, warehouses, streams. The plumbing that gets the right data to the right place.
open collective · AI for SMEs
From raw data to
business outcomes.
Alirion is a collective of practitioners helping small and mid-sized businesses adopt AI — sharing open tools, tested patterns, and hands-on guidance across the full stack, from raw data to running agents.
What we work on
ML systems
Models grounded in your data and deployed into your workflow — not a vendor portal.
AI agents
Authored in AML, reviewed by humans, run in production. Reviewable by product, legal, and compliance.
Governance
Guardrails, IAM, audit trails — applied uniformly across pipelines, models, and agents.
End-to-end
Collective knowledge, where you need it.
Most SMEs don't have the bandwidth to evaluate five vendors — a data team, a model shop, an MLOps platform, an agent vendor, a governance consultant. The collective brings together practitioners who've navigated this, sharing patterns and open tools across the full stack so you don't start from zero.
- Pipelines and warehouses, not just dashboards
- Models grounded in your data, not generic APIs
- Agents you can read, review, and roll back
- Governance applied uniformly across the stack
crm-lookup.tool.md
---
kind: tool
name: crm.lookupCustomer
auth: oauth:crm
---
# Description
Fetch a customer record by email
so an agent can act with context.
## Inputs
- email: string
## Returns
- id, plan, openTicketsOpen-source at the core
Agents you can read in a coffee break.
The agent layer is built on AML — our open-source format for describing AI agents and their dependencies. Plain Markdown for behavior, plain YAML for wiring. Reviewable by product, legal, and compliance — without an engineer translating.
- Markdown-first, YAML for structure
- Versionable in git like the rest of your code
- Framework-agnostic — bring your model
support-triage.agent.md
---
kind: agent
name: support.triageTicket
model: gpt-4o-mini
tools:
- crm.lookupCustomer
- helpdesk.createTicket
---
# Goal
Triage an inbound support request
and route it to the right team.
## Steps
1. Look up the customer by email
2. Classify the issue type
3. Assign priority and open a ticketHow a working system comes together
One community. The whole journey.
From the first data conversation to an agent in production — and the measurement that proves it worked. The collective covers every step, so you're never starting from zero.
Start from the outcome
It starts with the business result you're chasing — revenue, cost, risk, speed — not the technology.
Get the data flowing
Pipelines and warehouses that make your data trustworthy and ready to act on.
Make it learn
Models grounded in your data and pointed squarely at the metric that matters.
Put it to work
Agents that take action — authored in AML so your whole team can read and review them.
Prove the impact
Guardrails and measurement that tie every decision back to the outcome — then iterate together.
From business goal to working system.
Explore our open-source projects — including AML, a shared format for describing AI agents, and AML Studio where they're authored and reviewed.