An intelligence layer for the agents you already use
Meet the council behind your agents.
See how specialist agents, shared memory, and evidence support one decision you keep. Explore the system, then bring the method to the coding agent you already use.
- Human-held authority
- Inspectable routes
- Portable files
- 01 / Name the workA product lead asks for a weekly evidence brief and names the sources to check.
- 02 / Route specialistsOne role gathers sources. Another checks counterexamples and missing context.
- 03 / Return evidenceThe brief separates supported claims, open questions, and missing citations.
- 04 / Make the callThe product lead approves, revises, or holds publication. The key stays human.
A place to begin
Give your intelligence a memory.
Shape the context, character, and evidence your agents work with. Edit five plain files in Memory Studio, then carry them into your own workflow.
What comes back
What a useful run gives back.
A route earns trust when its output helps you decide. In this illustrative weekly product brief, these are the four artifacts worth asking your own agent for.
- 01
A bounded brief
The question, source window, exclusions, and decision owner are written down first.
- 02
An evidence ledger
Each claim points to a source or stays explicitly unverified.
- 03
A countercheck
A second role records contradictory findings and what it could not resolve.
- 04
A decision record
The owner approves, revises, or holds the brief with reasons attached.
Explore the intelligence layer
The work moves. The key stays with you.
Follow one bounded route through people, agents, memory, and proof. Select a point to see what it does.
Drag to explore · Select a point
The Starlight System
One instrument. Five layers.
Scroll to open the system you’ll build.
Every brief runs through all five. You stay the conductor.
Explore the workshopsFrom map to a working contract
Inspect the package before you install.
Agent cards show roles. A Department OS package also shows missions, evaluation gates, permissions, and rollback in one versioned contract.
First review candidate
Growth Studio Department OS
Research, produce, prepare, and learn from founder-led growth work with visible human authority.
5 missions · 5 roles · 5 evaluation gates
Inspect the package →Growth Studio is an unsigned review candidate whose implementation remains in private source. No public archive or self-service install path exists yet, and marketplace approval is never inferred from a local manifest.
The short answers
What you need to know.
Do I need a new AI tool?
No. Starlight is designed around the coding agents you already use. The start path shows what to bring into your own workflow.
Does the council make final decisions?
No. Published routes describe bounded roles and human approval points. A person keeps the irreversible call.
Is every department ready to install?
No. Growth Studio is an unsigned review candidate. Its page states what can be inspected and what is not yet distributed.
What you carry out
Plain files. Your decisions stay portable.
The core contracts are readable in any text editor. Runtime adapters need their own verification; the files stay yours to inspect, revise, and carry forward.
- 01AGENTS.mdRoles, authority, boundaries, and stop conditions.human-owned
- 02SKILL.mdA portable procedure for work worth repeating.tool-independent
- 03MEMORY_MAP.mdWhat persists, what expires, and what stays private.private by default
- 04EVAL_RUBRIC.mdThe evidence required before work moves forward.inspectable
- 05RUN_RECEIPT.jsonA decision-ready account of what happened and why.accountable
The workshop
Bring one real workflow. Keep the system behind it.
A focused working lab for founders and teams ready to turn an important workflow into a governed, reusable intelligence system — built on your own repos, running before you leave.