Documentation

Everything you need to get started with Qstar Labs.

Quickstart

Three steps to start building your workspace's knowledge base.

Step 1 — Create a workspace

$ npx qstarlabs login
# Sign in with OAuth or email
# Create or join a workspace

Step 2 — Connect to Claude Code

// Add to your Claude Code MCP config:
{
  "mcpServers": {
    "qstarlabs": {
      "command": "npx",
      "args": ["qstarlabs-mcp"]
    }
  }
}

Step 3 — Your first dream

# Finish a coding session. OpenCore captures a dream.
# Q* watches docs/dreams/ and uploads automatically.
# You're now contributing to your workspace's knowledge base.

MCP Tools Reference

Tool	Params	Returns
qstarlabs_search	query, limit?, min_confidence?	Gradient[]
qstarlabs_filter	substrate_type?, domain_tags?, stack_tags?, min_confidence?	Gradient[]
qstarlabs_related	gradient_id, relationship?	Gradient[]
qstarlabs_submit_dream	dream: Dream, idempotency_key?	{ ingestion_id, status }
qstarlabs_submit_delta_batch	delta_batch: DeltaBatch, idempotency_key?	{ ingestion_id, status }
qstarlabs_profile	(none)	{ reputation_score, dreams_submitted, convergence_rate }

REST API Reference

Method	Endpoint	Description	Auth
POST	/api/v1/workspaces/:ws/dreams	Submit a dream	contributor+
POST	/api/v1/workspaces/:ws/delta-batches	Submit a delta batch	contributor+
GET	/api/v1/workspaces/:ws/gradients/search	Semantic search	reader+
GET	/api/v1/workspaces/:ws/gradients	Structured filter	reader+
GET	/api/v1/workspaces/:ws/gradients/:id/related	Graph traversal	reader+
GET	/api/v1/workspaces/:ws/profile	Contributor profile	reader+
POST	/api/v1/workspaces/:ws/gradients/:id/rollback	Rollback gradient	admin+

Full API documentation coming soon. Auth via API key in Authorization: Bearer <key> header.

Concepts

Dream: A session reflection produced by OpenCore's dream skill. Contains a pattern, observation, context, and anti-pattern — never code or file paths. Uploaded to Q* automatically.
Gradient: A workspace knowledge entry promoted after cross-project convergence and eval-gate validation. Living entity with confidence score, version tracking, and graph relationships.
Convergence: When N independent dreams from M distinct projects arrive at the same pattern. The mechanism that separates workspace knowledge from individual opinion.
Substrate (S1-S6): The six scaffold layers from the Frontier and Localhost paper: Instructions (S1), Skills (S2), Memory (S3), Tools (S4), Orchestration (S5), Governance (S6).
Reputation Score: A 0.0-1.0 score derived from convergence rate (50%), consistency (30%), and diversity (20%). Determines how much weight your dreams carry in convergence calculations.
Candidate Pool: Where distilled dreams wait before promotion. Patterns stay here until convergence thresholds are met. A single brilliant insight waits for independent confirmation.