Software
Coordinate product, engineering, data, design, and operations before a change ships.
cptodevdata-engineerdevopsproduct-designpollm-engineer
$agents-city demoHow to solve real work with agent cities.
The same system organises code, knowledge, and coordinator agents in software, healthcare, legal, finance, marketing, sales, research, operations, or a domain of your own.
Start with the vocabulary of your work.
Each pack supplies evidence criteria and suggested roles as editable Markdown. It grants no authority and installs no skills. blank and custom let you start without a hidden profile.
Healthcare
Bring clinical direction, patient safety, operations, data, and compliance into one bounded decision.
clinical-directorclinicianpatient-safetyclinical-opshealth-datahealth-compliance
$agents-city demo --domain medicinaLegal
Work a matter across deadlines, precedent, client context, compliance, and knowledge.
managing-partnerassociatecomplianceknowledgeops
$agents-city demo --domain legalFinance
Coordinate close, controls, analysis, compliance, and operational cycles.
cfocontrollerfin-analyticsopscompliance
$agents-city seat --domain financeMarketing
Align brand, content, acquisition, lifecycle, design, and measurement.
brand-leadcontentperformanceseolifecycleproduct-design
$agents-city seat --domain marketingSales
Connect account evidence, revenue operations, enablement, and customer outcomes.
revenue-leadaccount-executiverevopscustomer-successenablement
$agents-city seat --domain salesResearch
Separate methods, evidence, ethics, knowledge, and verification before synthesis.
research-directorresearchermethodsresearch-opsethicsknowledge
$agents-city seat --domain researchOperations
Resolve dependencies across programmes, processes, quality, knowledge, and delivery.
operations-leadprogram-managerprocess-ownerqualityknowledge
$agents-city seat --domain operationsYour domain
Build a city around your own team, vocabulary, and way of deciding.
city-leadspecialistqualityknowledgeblank
$agents-city seat --domain customStart with the problem, not the command.
Every guide models a reproducible situation: what stays on each owner’s machine, which seats may talk, what evidence each agent owns, and how to check the result.
For companies
Teams, clients, suppliers, and specialists connecting their cities without sharing machines or diluting authority.
01
Company · product team
02How to coordinate AI agents across multiple repositories
Give API, web, and QA separate evidence roles while one chair owns the release goal and the final decision.
Company · engineering and product
03How a developer’s agent can ask Product before building
Let Engineering keep moving while Product or Business answers through an agent grounded in the owner’s criteria, knowledge, authorised data, and existing tools.
Company · agency or consultancy
04How to isolate AI agent work between client projects
Give every client a separate city and open a road only when two chairs have a legitimate reason to exchange information.
Company · operations leadership
How to run an AI agent committee for a business decision
Compare independent evidence from operations, finance, customer success, and compliance without turning the process into a vote.
For individuals
Solo builders, researchers, and multi-model users working from their own machine.
01
Individual · solo builder
02How to manage a solo software project with several AI agents
Run specialist repository agents without becoming the human message router between unrelated terminals.
Individual · researcher or analyst
03How to compare research with AI agents without groupthink
Collect independent positions from literature, methods, data, and ethics before one research chair sees the full comparison.
Individual · multi-model user
How to use Claude, Codex, OpenCode, and Kimi in one project
Choose the best runtime for each window without forcing every provider through the same terminal fallback or losing one coordination boundary.
Quick answers for configuring and operating a city.
These recipes complement the complete guides when you already understand the problem and only need the correct CLI flow.
Start with one city and no provider account
Use it when
You want to understand the product before choosing a runtime or touching your repositories.
Flow
Choose the software, healthcare, or legal disposable story.
Follow the chaired committee, failed verification, replan, and corrected close in the live Hall.
Stop it with Ctrl-c; the temporary city disappears.
$agents-city demo$agents-city demo --domain medicina$agents-city demo --domain legalUse Codex as the city chair
Use it when
You want Codex to own the goal and chair repo evidence while other windows use any supported runtime.
Flow
Create or open the city.
Choose engines and set Codex for the seat.
Open the tmux session and verify the configured windows.
$agents-city setup --city product$agents-city seat --city product --engines$agents-city seat --city productMix runtimes by repository
Use it when
Different repositories benefit from different providers or a local model.
Flow
Open the engine picker.
Assign a runtime and optional model to each window.
Keep persistent choices in the owner card; use --model only for a one-launch override.
$agents-city seat --city product --engines$agents-city seat --city product --model gpt-5 --effort highKeep client work isolated on one machine
Use it when
Product, client-a, and research must keep separate goals, knowledge, repos, and road allowlists.
Flow
Create each city independently.
Select and operate one without starting the others.
Connect only the two seats that need to exchange information.
$agents-city cities create product$agents-city cities create client-a$agents-city road connect product client-aMake one release decision across several repos
Use it when
API, web, and QA evidence can materially change whether you ship.
Flow
Open a brief with only relevant participants.
Collect isolated positions and synthesise conflicts.
Decide with executor and verifier.
Close only after independent verification passes.
$agents-city committee schema open$agents-city committee open --input release.json$agents-city committee show <deliberation-id>Connect two local cities
Use it when
Two autonomous work domains on the same machine need a controlled route.
Flow
Create both cities.
Connect them symmetrically.
Open both seats and inspect their roster.
$agents-city road connect product research$agents-city road list product$agents-city bus rosterConnect seats on different machines
Use it when
A collaborator owns another city, account, and machine.
Flow
Self-host the narrow remote relay.
Have the relay operator mint one owner credential per machine.
Exchange public invitation files.
Each side accepts the other independently.
$agents-city road invite product > product.invitation.json$agents-city road connect product research.invitation.jsonOpen only part of a large city
Use it when
The city owns many repos but today’s work needs only API and web.
Flow
Keep the full repo set on the owner card.
Use a one-launch filter without changing persistent configuration.
$agents-city seat --city product --only api,webInspect repo skills without installing anything
Use it when
The chair needs to know which repo agent already has a relevant capability.
Flow
List recognised skills.
Choose the agent by role, ownership, scope, and live skills.
Let that runtime decide whether to invoke its own skill.
$agents-city skills product