What is Constellation
Depending on how much time you have.
10 seconds
Your institution’s rules, enforced at the moment of action — not in a quarterly review afterwards.
30 seconds
Every organisation has rules about what it can and can’t do — spending limits, approval requirements, topics that are off-limits, commitments that need sign-off. Today those rules live in documents, policies, and people’s heads.
Constellation makes those rules structural. When an AI agent or person is about to do something consequential — send an email, commit money, publish content — the system checks it against your institution’s rules before it happens. If it crosses a boundary, it tells you which rule, why, and who to escalate to.
Every action is traced. Setup takes minutes. The system never blocks — it informs. The human always decides.
2 minutes
Constraint enforcement
Constellation supports seven constraint types: authority (“who can approve this?”), threshold (“how much?”), prohibition (“never do this”), timing (“not until 48 hours after the board meeting”), domain-topic (“no content touching active litigation”), audience (“board members only”), and sequence (“legal review must come first”).
For AI agents, constraint checking happens automatically via MCP integration. For humans using AI tools, the same constraints apply through the web interface or API. One set of rules, enforced regardless of who or what is acting.
Contestation
Most governance tools assume compliance. Constellation assumes disagreement. Any constraint can be formally challenged through the Forum. Challenges require standing and evidence. Adjudicators issue rulings. Rulings carry precedent weight that influences future decisions. If you disagree with a ruling, you can contest it. If the contestation fails, you can appeal to a final arbiter. This is a common law system inside a product.
Measurement
Every check, escalation, and decision generates data. Constellation computes a continuous Governance Coordination Index across five dimensions: authority clarity, escalation efficiency, documentation proportionality, coordination velocity, and structural stability. Governance becomes a number you can track over time — not a feeling.
What it’s not
Not compliance. Compliance checks what happened after the fact. Constellation checks before the action happens.
Not advisory. Constellation doesn’t advise, optimise, or decide. It enforces the institution’s own commitments.
Not surveillance. Constitutionally prohibited from ranking individuals, scoring compliance, or enabling trace-based discipline.
10 minutes
Four governance layers
Constellation is not a single tool. It’s a governance architecture with four interconnected layers:
Charter
Decisions, commitments, constraints, escalation chains, authority roles. The institutional record.
Knowledge Graph
Institutional memory structured as a navigable graph. Concepts, relationships, sources. Feeds into constraint checking and decision support.
Forum
Three-layer dispute resolution: challenges, contestations, appeals. Standing requirements, evidence, rulings, precedent library.
Observatory
External signal monitoring across 9 domains. Drift detection from baselines. Alerts when governance needs to adapt.
Progressive trust for AI
AI delegation is not a switch. In Shadow Mode, AI analyses events and recommends actions while the institution decides. Constellation tracks agreement rate per decision category. When accuracy exceeds the threshold with sufficient sample size, delegation becomes available — and can be suspended instantly with full rollback. A Guardian system activates human override when the AI encounters situations outside its calibrated range.
Built for AI agents
Constellation ships as an MCP server that plugs into any agent supporting the Model Context Protocol. Eight tools: check, boundary, record, escalate, preview, query, status, and create_constraint. Cached for speed (<200ms), circuit breaker protected for resilience.
This means every Claude Code session, every AI SDR, every automated workflow can be governed by the institution’s own rules — without changing the agent’s code.
Common questions
“Is this just compliance software?”
No. Compliance checks what happened after the fact. Constellation checks before the action happens. It’s the difference between a security camera and a lock.
“Does it block actions?”
Never. Constellation informs. The human always decides. But the decision is made with full institutional context, not from memory. And the override is recorded — you can choose to break the rule, but you must do so consciously.
“What if I disagree with a rule?”
Challenge it through the Forum. You need standing and evidence. An adjudicator issues a ruling. That ruling carries precedent. If you disagree with the ruling, you can contest it. If that fails, you can appeal to a final arbiter. This is governance with due process, not top-down control.
“How is this different from better system prompts?”
System prompts are per-agent. Constraints are per-institution. When you have 20 agents, you can’t maintain 20 system prompts. When rules change, you update one constraint — not 20 configurations. And system prompts have no audit trail, no contestation, and no institutional memory.
“We’re only three people. Do we need this?”
If your three people are running AI agents that send emails, make commitments, or handle money — yes. The constraints aren’t about headcount. They’re about how many things are happening that cross institutional boundaries. Three people with 20 agents have the same governance problem as 200 people with none.
See where your governance stands
20 questions. 5 minutes. Free. No sign-up required.