SymphonyAgent Orchestration for Software Development

Assign tickets to AI agents. They research, design, implement, and open PRs — while you stay in control.

Architecture

Three cooperating processes share one SQLite DB — web UI, orchestrator daemon, and ephemeral MCP servers. Understand how it all fits together.

Read the architecture

Phase Pipeline

Issues mature through research → architecture → grooming → ready before a worker touches code. Agents validate their own artifacts at each gate.

How phases work

Agent Types

Worker, Judge, Planner, Researcher, Scanner — each with a distinct role. Workers implement; Judges review; Researchers analyse; Scanners discover.

Meet the agents

MCP Tools

20 tools give agents structured access to update issues, record findings, create PRs, and manage pipeline state — all over stdio.

Browse the tools

Configuration

Tune timeouts, concurrency, retry backoff, model selection, and per-project agent toggles. Includes a tuning guide for small, large, and cost-sensitive projects.

Configure Symphony

Prompt Engineering

Deep reference for how prompts are assembled — layer order, token budgets, phase contracts, custom instructions, and the learning injection system.

Prompt reference

See It Running

Project kanban board showing issues across Research, Design, Grooming, Todo, and In Progress columns

Project kanban board — issues flow through phase columns before reaching implementation

Pipeline view showing phase distribution and metrics

Pipeline view — phase distribution, time-in-phase metrics, and artifact tracking per ticket

Workspace overview showing all projects with pipeline health

Workspace overview — all projects, pipeline health bars, and live agent count

Agents view showing per-project agent toggles and running agents

Agents view — enable/disable worker, judge, planner, researcher, and scanner per project