← Back to Category

AI Coding

synth-city

Synth City, formerly Claude Deck, is a self-hosted local control plane for coordinating coding-agent teams on GitHub issues through to reviewed pull requests, with a dashboard for progress, decisions, and human oversight.

coding agentsGitHubself-hostedteam coordinationdeveloper tools

Directory information is for reference. Check the official site for current features, pricing and availability. Listing does not imply hands-on testing or a purchase recommendation.

About this product

Synth City keeps GitHub as the source of issues and pull requests while a local Deck controller observes and intervenes in bounded coding-agent attempts.

A factory is a supervisor, the Deck controller, and a team working a GitHub backlog, with a human Director owning the outcome and retained decisions.

Key Features

  • Overview, Work, and Repositories show complete filtered work counts, owners, waiting reasons, PR context, and watched-scope intake.
  • Harnesses provides an operations and readiness catalog for implemented providers with guarded native pages.
  • A config editor browses, inspects, and edits Claude Code JSON settings or Codex TOML settings, including profiles and feature flags.
  • MCP Servers can be added, edited, tested, and managed with OAuth support, registry browsing, and views of tools, resources, and prompts.
  • Agent Bridge discovers and monitors agent tmux sessions and can attach up to four terminals in a 2x2 grid with per-pane controls.
  • Agent Mail coordinates local agents through durable per-repo identities, structured context requests, handoffs, and a team mailbox.
  • Agent Teams saves reusable rosters of agents and launches or reuses their sessions.

Use cases

  • Deliver work from a GitHub issue backlog with a Leader directing implementers and verifiers toward reviewed pull requests.
  • Inspect owners, waiting reasons, and PR context in the dashboard before approving a protected remedy.
  • Coordinate mixed teams across Claude Code, Codex CLI, GitHub Copilot CLI, OpenCode CLI, and Pi sessions.

Who it is for

  • Technical operators with existing repositories, local agent CLIs, native toolchains, and their own model credentials.
  • Teams that mix harnesses and need to keep human decisions in the loop.

Requirements

  • Python 3.11+
  • Node.js 22.12+
  • At least one supported local agent CLI on the same host: Claude Code, Codex CLI, GitHub Copilot CLI, OpenCode CLI, or Pi
  • Linux for agent-team pane binding

Quick Start

  1. Clone the repository with git clone https://github.com/adrirubio/claude-deck.git and change into the claude-deck directory.
  2. Run ./scripts/install.sh for the native install path.
  3. Run Deck in the same environment where agent CLIs and credentials are installed, with at least one supported CLI present.

Limitations and considerations

  • Docker is not supported because containers cannot see host-installed CLIs, tmux sessions, or native agent credentials.
  • Finished is a tracking state, not independently reviewed delivery, reliability, cost, or time saved.
  • Cost is unknown, and configuration checks do not verify model access.

License and links

Code license: MIT

Privacy and data

  • There is no Deck cloud account and no Deck telemetry sent elsewhere; GitHub and native CLIs use their configured services and credentials.
  • Deck reads and writes real local agent configuration files, so UI changes affect actual files and backups are advised before major edits.

Sources

Source review: 2026-11-12