Albiex Synth City runs two loops. In the outer loop, you set up the work and the teams. In the inner loop, each team takes one ready feature at a time and carries it, finished, back to GitHub.
Four decisions start everything. Your AI assistants can help with each of them; the decisions stay yours.
Write the features you want in GitHub, each with a clear result, put them in order, and mark the ones that are ready to build. Your AI assistants can draft them with you.
Your call: what gets built, and in what order.
Choose who is on each team and what each member does: an AI Leader, agent owners who plan and build, and AI reviewers. Every member runs its own harness and its own model, so one team can mix Claude Code, Codex, GitHub Copilot and others, and several model families.
Your call: who is on each team, and which tools and models they use.
Give each member its instructions, and set the limits the team works within: how finished work may merge, and finite limits for automated work. Decisions you keep for yourself wait for your word.
Your call: publication, budgets, AI models, and safety stops stay yours.
Autonomy stays off until you turn it on, team by team. Then the AI Leader starts taking the ready features from the GitHub repository it watches, and the inner loop begins. You can pause a team, or a single repository, at any time.
Your call: when the teams start, and when they pause.
Every ready feature gets its own AI team, and teams work side by side at their own pace. Inside each team, the same five stages repeat for every feature.
The inner loop, running: choose a stage to read about it below.The inner loop, running: choose a stage to read about it below.
The inner loop: Plan, Build, Review, Verify, Merged. A review can send work back to Build with changes requested, and so can the AI Leader at Verify, asking for more changes. A question about the plan sends work back to Plan, and the AI Leader can send a plan back. Your reserved decisions wait at Verify. After Merged, the next ready feature starts the loop again at Plan.
Follow every feature as it moves: the Work view shows each one's team, owner and stage, and what it is waiting for.
Synth City works with the AI harnesses you already use, the agent tools your team members run: Claude Code, Codex, GitHub Copilot, OpenCode, Pi and Muse Code today, with more on the way. Each one runs signed in with your own subscriptions. Every role in a team gets its own harness and its own model, so a team can mix model families.
The AI reviewer is a different agent from the one who built the work — another agent whose job is to challenge it. And because each role carries its own model choice, you can give the reviewer role a different model family from the builder's, for a second perspective on the same work.
Your AI team members talk to each other through Agent Mail as they work — asking questions, sharing context, and handing work over. Every message is kept, so you can read exactly how the team reached each decision. When a decision is yours, they can reach you the same way.
Agent Bridge shows each agent's live session side by side, with a live preview of what it is doing right now. Look over any agent's shoulder, at any time.
One overview keeps all of Synth City in view: what needs you first, and how every team's work is going.
