Hosted by 
This is a multi-step software factory intensive to give participants tools and principles to configure complete software factories for their use case. Participants will learn how to think about the dimensions of configurations for software factories and how to elevate their software development infrastructure above individual coding agents.
The curriculum runs over two days. Day 1 stands up a working factory and takes it apart to see how it is made: packs, agents, formulas, orders, and the five ways agents coordinate. Day 2 points that machinery at a project of your own and extends it, from a menu of options you choose between rather than a fixed sequence.
You start from a base factory that already works, built on the out-of-the-box examples that ship with Gas City and two reference agents from Gas Town, mol-polecat-work and mol-refinery-patrol. What you layer on top is your choice: GitHub pull requests and quality gates are already in the base, and the options add specialized reviewer agents, an architecture-best-practices loop, a multi-model review, a gate at the front of the queue, and a loop that lets the factory propose changes to itself.
Gas City is an open-source framework for running multi-agent systems. It abstracts the primitives of multi-agent coordination — agents, packs, rigs, beads, sessions, orders, routes — so that any multi-agent architecture can be expressed within the same framework rather than re-invented each time. Once you've learned the primitives, you can swap out the specific agent roles and build pipelines for code review, research, data processing, ops automation, etc.
For the authoritative definitions of every Gas City term (agent, pack, rig, bead, sling, order, route, formula, overlay, etc.), see the Gas City glossary. Brief term map:
| Gas City Term | Analogous Term |
|---|---|
| bead | Issue / ticket / task |
| convoy | Epic / batch |
| dog | Daemon / cron worker |
| formula | Workflow / pipeline / recipe |
| Message / inbox item | |
| order | Cron job / scheduled task |
| pack | Plugin / module / package |
| rig | Workspace / repository |
| sling | Job dispatch / enqueue |
Stuck on a step, want to share what you've built, or looking to collaborate with other participants? Join the Actual AI User Community Slack to ask questions and get help from other participants.
You’ll need at least one coding agent subscription (or API key) to run the workshop. Having more than one gives you broader capabilities and redundancy.
- Recommended: Claude Code Max (20x) or Codex Pro (20x)
- Minimum: Claude Code Max or Codex Pro (standard tier)
- Alternatives: Gemini CLI, OpenCode (compatibility not guaranteed)
To prepare for the workshop, please install the Actual Factory Demo and follow the instructions to get it running. This proves your machine is ready to go and gives you a hands-on understanding of the basics of Gas City.
Two days. Workshops (W) are instructor-led, while labs (L) are participant-led.
| Session | Type | Duration | Title | Page |
|---|---|---|---|---|
| W1 | Workshop | 30 min | Vocabulary and Concepts | W1 |
| W2 | Workshop | 45 min | Cloud Box and Preflight | W2 |
| W3 | Workshop | 60 min | Run Your Factory | W3 |
| W4 | Workshop | 60 min | Tour the Factory | W4 |
| W5 | Workshop | 30 min | Observability and Traceability | W5 |
| L1 | Lab | 45 min | Point the Factory at Your Own Project | L1 |
| L2 | Lab | 30 min | Capability Map (starts) | L2 |
| Session | Type | Duration | Title | Page |
|---|---|---|---|---|
| L2 | Lab | 30 min | Capability Map (finishes) | L2 |
| L3 | Lab | 45 min | First enhancement — one of the six options | L3-L5 |
| W6 | Workshop | 60 min | Advanced Concepts | W6 |
| L4 | Lab | 60 min | Second enhancement | L3-L5 |
| W7 | Workshop | 45 min | Sharing Your Factory | W7 |
| L5 | Lab | 45 min | Third enhancement | L3-L5 |
The three feature labs draw on the same menu of options. In L3 you pick one of these six, then in L4 and L5 you can either choose one you have not used yet, or a feature of your own from your capability map.
| Option | Solves |
|---|---|
| Bead creation formula extensions | Beads arrive with no design, test or docs thinking attached |
| Specialized domain reviewers | One reviewer judging everything reviews nothing well |
| Architecture best-practices loop | Reviews are opinions rather than scores against named principles |
| Strengthen the review system | One model is one point of view, and it is confidently wrong sometimes |
| Self-improvement loop | The factory produces evidence about itself and nobody reads it |
| Bead gate checks | Work reaches an implementer under-specified and it has to guess |
W3 installs one pack, base-factory, and gets a working multi-agent factory with an example of every Gas City primitive already running in it: formulas, an agent persona, orders, a rig, and mail. Day 2 extends it.
The four appendix pages assemble that same factory by hand, step by step. They are off the taught path and worth reading when you want to know why a layer exists rather than what it does.
Ready? Start at W1 Vocabulary and Concepts.