Runner Teams & Multi-Agent Pipelines

From Single Runner to Dedicated Teams

Until now, AI runners in Spedy operated as a single pipeline: move a ticket to a status column, a runner picks it up, done. Simple, but limited.

With Runner Teams, you can now define dedicated teams — each with their own pipeline stages, prompts, tool permissions, and status configuration. Assign a team to a ticket, and the team's pipeline takes over.

What's New

Runner Teams are configurable units that define how AI processes a ticket:

• Pipeline Stages — Each team has its own ordered sequence of stages (e.g., Planning, Coding, Review). Every stage has a custom system prompt, role (Planner or Executor), and allowed tools.
• Status Configuration — Teams can define their own working, completion, and failure statuses. When the AI starts working, the ticket moves to the team's working status. When it finishes, the ticket moves to the completion status.
• Direct Assignment — Instead of relying on a trigger column, you assign a team directly to a ticket. This gives you precise control over which AI configuration handles which work.

Agent Team Templates

For stages that need more than a single agent, you can assign Agent Team Templates — reusable multi-agent configurations that define how multiple AI agents collaborate within a single stage.

Templates support three execution strategies:

• Parallel — Independent agents run simultaneously (e.g., security review + quality review + test validation at the same time)
• Sequential — Agents run in order, each receiving the previous agent's output (e.g., plan → implement → review)
• Fan-out & Merge — Parallel workers feed into a final lead agent that synthesizes results

Spedy ships with four built-in presets:

1. Code Review — Security reviewer, quality reviewer, and test validator run in parallel, then a review lead synthesizes and applies fixes
2. Implement & Review — Sequential planner → implementer → reviewer pipeline
3. Full Stack Team — Architect plans, then frontend dev, backend dev, and test engineer work in parallel, with a final reviewer
4. Bug Investigation — Three investigators explore different hypotheses in parallel (data flow, edge cases, environment), then a synthesizer determines root cause and applies the fix

You can also build fully custom templates with your own agents, prompts, tools, and dependency graphs.

Job Lifecycle

When you assign a team to a ticket:

1. A job is created in READY state with the team's pipeline configuration
2. The next available runner picks up the job
3. The runner executes each pipeline stage in order
4. On completion, the ticket moves to the team's completion status

If you unassign a team from a ticket, any active jobs are automatically cancelled — the runner receives a stop command and the job transitions to STOPPED.

Getting Started

1. Go to Settings → Runners → Runner Teams
2. Click Add Team — a new team is created with default Planning + Coding stages
3. Click on a team to configure its stages, prompts, and status mapping
4. Go to Settings → Runners → Agent Templates to create or customize multi-agent team templates
5. Assign agent templates to individual pipeline stages for multi-agent execution

What's Next

Runner Teams lay the groundwork for more advanced AI workflows: conditional stage execution, approval gates between stages, and cross-team orchestration. Stay tuned.