Starlight Intelligence Lab

Human and agent platform

One contract across skills, plugins, APIs, and MCP.

The website does not maintain a marketing model and a separate agent model. Humans and agents inspect the same package graph, runtime state, evaluation criteria, and human-only boundary.

Public interfaces

Small surface. Explicit semantics.

The first release is intentionally read-only. It lists package truth and compiles deterministic plans. Authentication and external mutations arrive only with a concrete product need and a separate authority model.

GETpublic

/api/v1/departments

List release candidates and research previews.

GETpublic

/api/v1/departments/growth-studio

Read the complete public package contract.

POSTpublic

/api/v1/recommend

Compile product fit from five closed-domain inputs.

POSTpublic

/api/v1/install-plan

Compile a runtime-specific distribution, verification, and rollback plan.

POSTpublic

/mcp

Use the same compiler over Streamable HTTP MCP.

GETpublic

/.well-known/agent.json

Discover the public interfaces and authority boundary.

curl https://starlightintelligence.ai/api/v1/install-plan \
  -H "content-type: application/json" \
  -d '{"slug":"growth-studio","target":"codex"}'
MCP authority

All four MCP tools are closed-domain, deterministic, read-only, non-destructive, and idempotent. The endpoint cannot publish, purchase, access credentials, or mutate an external system.

Open formats

Portable by specification, honest by adapter.

The core package follows Agent Skills and Agent Plugins. The remote tool surface follows the current MCP revision while preserving a bounded legacy handshake path for existing clients.