Linear Integration
Polls a Linear project for candidate issues. Manages state transitions, respects priorities, and handles dependency graphs automatically.
Kata Symphony
Headless agent
orchestrator
Polls Linear for issues, dispatches parallel agents, manages the full PR lifecycle. Unattended, from Todo to merge.
symphony WORKFLOW.md --port 8080 v0.1.0 · Rust · Open Source
Autonomous ticket
execution engine
Parallel Agents
Configurable concurrency with per-state slot limits. Multiple agents work on different issues simultaneously across isolated workspaces.
Multi-Turn Sessions
Agents continue on the same thread across turns, preserving conversation history. Context carries forward through review cycles.
Full PR Lifecycle
Agents create PRs, address review feedback, resolve comment threads, and merge approved PRs. The complete cycle runs without intervention.
Live Dashboard
HTTP dashboard with running sessions, token tracking, retry queue, and polling stats. JSON API for programmatic access. Auto-refreshes every 2 seconds.
Workspace Strategies
Clone-local with hard links for speed, clone-remote for full isolation, or worktree for lightweight branching. Auto-cleanup on terminal state.
From Todo to merge
01
Todo
Human queues issue in Linear
Human 02
In Progress
Agent implements in isolated workspace
Agent 03
Agent Review
Agent addresses PR review comments
Agent 04
Human Review
PR awaits human approval
Human 05
Merge
Agent merges approved PR
Agent Human can set Rework at any point to restart from scratch
Live dashboard
3 Running
1 Retry
7 Claimed
12 Done
IssueStateElapsedWorker
KAT-142 In Progress 4m 32s local-0
KAT-138 Agent Review 12m 08s local-1
KAT-145 In Progress 1m 15s local-2
Common Questions About Kata Symphony
How do I install Kata Symphony?
Symphony is a Rust binary. Clone the repo and build with
cargo build --release. The binary lives at target/release/symphony. Run it with a WORKFLOW.md config file and a Linear API key.What agent runtimes does Symphony support?
Symphony dispatches to Codex (OpenAI's coding agent) via the
codex app-server interface. The agent config in WORKFLOW.md controls the command, stall timeout, and approval policy.Can Symphony manage multiple projects?
Yes. Run one Symphony instance per project, each with its own WORKFLOW.md. Different projects can use different workflow configurations, workspace settings, and prompt templates. Use different ports if running multiple instances.
How does the ticket lifecycle work?
Issues flow: Todo (queued) → In Progress (agent working) → Agent Review (addressing PR feedback) → Human Review (waiting for approval) → Merging (agent merges) → Done. Humans can also set Rework to restart fresh.
What happens when an agent fails?
Failed sessions enter a retry queue with configurable backoff. The dashboard shows retry timing and error details. If retries exhaust, the issue stays visible for manual triage. Stall detection catches hung agents.
How is Symphony different from Kata Orchestrator?
Orchestrator is a developer-facing harness that you interact with inside your coding agent (Claude Code, Gemini CLI, etc.). Symphony is a headless daemon that runs unattended, polling Linear and dispatching agents without human involvement. Orchestrator is hands-on; Symphony is hands-off.
Start orchestrating
Clone the repo, configure your WORKFLOW.md, and let Symphony handle the rest.
cargo build --release