Pipeline & Goals

Build goals

Every build starts by choosing a goal. Goals control which agents run and how deeply each phase is executed.

Goal	Use when	Agents	Typical time
prototype	Exploring an idea	~15 core agents	3–5 min
mvp	Building something real	Core + specialists	8–12 min
production	Shipping to users	All 58+ agents	15–25 min

Pipeline v2 (streaming)

Start a build with a POST to /pipeline/v2/stream:

{
  "prompt": "Build a FastAPI CRUD app with PostgreSQL and JWT auth",
  "goal": "mvp",
  "mode": "plus",
  "hitl_enabled": true,
  "hitl_phases": ["planning"],
  "project_name": "my-api"
}

The response is a text/event-stream. Each agent emits events as it runs:

event: phase_start
data: {"phase": "planning", "agents": ["IdeationAgent", "ArchitectureAgent"]}

event: agent_output
data: {"agent": "IdeationAgent", "content": "...", "tokens": 412}

event: checkpoint
data: {"checkpoint_id": "chk_a1b2", "phase": "planning", "type": "hitl"}

event: build_complete
data: {"project": "my-api", "files": 47, "lines": 3820, "ctq_score": 84}

Pipeline phases

Phase	Description
planning	Ideation, architecture, scope definition
design	Database schema, API contracts
implementation	Backend, frontend, infrastructure code
testing	Unit, integration, security, performance
release	CI/CD, deployment config, documentation

HITL — pausing at checkpoints

Set hitl_enabled: true and specify which phases to pause at:

{
  "hitl_enabled": true,
  "hitl_phases": ["planning", "release"]
}

When the pipeline hits a HITL phase, it emits a checkpoint event and pauses. Resume with:

POST /pipeline/v2/checkpoint/{checkpoint_id}/resolve

{
  "action": "approve",
  "feedback": "Architecture looks good, proceed"
}

action can be approve, reject, or modify. Use modify with a feedback string to redirect the agent before it continues.

Session management

# List all your sessions
GET /pipeline/v2/sessions?status=running

# Get full session details and all agent outputs
GET /pipeline/v2/session/{session_id}

# Cancel a running build
DELETE /pipeline/v2/session/{session_id}

Modes

Mode	Description
standard	Uses the default LLM routing (MoE)
plus	Routes all agents to the highest-capability models

Use plus for Production goal builds where quality matters most. standard is faster and cheaper for prototypes.