Govern every decision
your organisation makes.
Your organisation runs on decisions - made in Slack, email, and conversations, then scattered and forgotten. Constans captures them where they happen, governs them with human confirmation, and answers from them every time after.
No new system to roll out. No behaviour change required.
The problem
Decisions live everywhere.
Nowhere holds them.
"We made that call last quarter. Why are we discussing it again?"
Decisions get debated, resolved, then quietly forgotten when the people who made them move on or move out. The same conversation comes back six months later, with nobody confident enough to point at the answer.
"The person who decided that has left. Nobody remembers the reasoning."
Institutional memory walks out the door with key hires. Reasoning becomes lore, then myth, then nothing. New people inherit the conclusion without the context that justified it.
"Our internal AI confidently answered with something we never sanctioned."
AI assistants are arriving inside organisations faster than the governance to anchor them. Confidence without provenance is exposure - to customers, regulators, and your own audit committee.
Pure automation is too risky for the decisions your organisation has to live with. So Constans is collaborative AI: the system proposes, your people confirm, and every answer your organisation gives is one a human has stood behind.
The system
A governance brain that sits
inside your org.
Constans takes the decisions your team sends it - from the web app, in Slack, by email, as document uploads, or through MCP - and turns them into a governed body of decisions. When you're ready, optional mailbox monitoring catches the decisions made in normal correspondence too. Scoped to the whole org, a department, a project, or a single team. Sentinel watches the whole picture, surfacing patterns that cross those boundaries.
Ingestion
Built around what your team sends to it.
- The Constans web app
- Slack via @constans mention or slash command
- Per-scope email addresses
- Uploaded documents and meeting notes
- AI agents and tools via MCP
Optional, when you're ready: passive monitoring of designated Workspace or M365 mailboxes.
Governance
Signals become governed knowledge.
- Signals with governance shape are distilled into proposed decisions
- The right person reviews and confirms
- Confirmed decisions become canon
- Scoped: org, department, project, or team
Reasoning
Anyone with permission can ask.
- Ask in Slack, by email, in the web app, or through your chosen AI platform
- Every answer cites its source decisions
- The AI distinguishes confirmed canon from proposed decisions, and surfaces the status of every claim
- Provenance is built in, not bolted on
The loop
A question is asked.
If Constans has validated knowledge, it answers.
If not, it identifies who knows, asks them inside their normal work, and incorporates the answer.
The gap is closed once, then forever.
Query
A question enters the system.
Answer
If governed knowledge exists, respond immediately.
Detect gap
If knowledge is missing, identify the gap precisely.
Route to expert
A structured question is sent to the right person.
Capture and validate
The answer is validated, versioned, and added to the knowledge base.
Answer next time
The same question is answered instantly, from governed knowledge.
Query
A question enters the system.
Answer
If governed knowledge exists, respond immediately.
Detect gap
If knowledge is missing, identify the gap precisely.
Route to expert
A structured question is sent to the right person.
Capture and validate
The answer is validated, versioned, and added to the knowledge base.
Answer next time
The same question is answered instantly, from governed knowledge.
The core tenet
AI proposes.
Humans decide.
Every decision Constans surfaces is a proposal. A human confirms it before it becomes canon. Canon is the only thing the system will answer authoritatively from.
The governance is not a feature. It is the design.
Sentinel - Continuous oversight
Policy drift doesn't announce itself.
Most knowledge systems wait to be asked. Sentinel watches the body of decisions as it grows and raises its hand when something needs a human's attention, before anyone thought to look.
Conflict
When two answers can't both be right.
Two teams settle the same question different ways. Constans catches the contradiction and surfaces it to the scope that owns the call - before it reaches a customer, a regulator, or a board paper.
Drift
When practice quietly drifts from policy.
Decisions diverge from agreed policy one small call at a time, until the gap is structural. Constans flags the drift to the people accountable for the policy - while it's still a correction, not a crisis.
Convergence
When the same answer is found five times over.
The same conclusion gets reinvented in different corners of the org. Constans spots the pattern and surfaces it for promotion into shared canon - so it's decided once, for everyone.
Surfaced like this.
Each alert lands with a category, a confidence score, and the evidence the system used to raise it. The human accountable for the scope decides what happens next - acknowledge, propose a decision, or suppress.
Risk accumulation
Multiple small issues compounding into a structural concern at the scope above them.
Policy drift
Decisions diverging from a stated parent constraint or guardrail, one small call at a time.
Coverage gap
A required control - security review, audit, sign-off - that has not been evidenced.
Emerging pattern
A signal forming across decisions that sits below the threshold for conflict or drift, but is consistent enough to surface for early review.
Real alerts from a synthetic organisation.
Constans surfaces. The call always stays with the human who owns it.
Morning briefing
Leadership wakes up informed.
Every morning, each scope leader receives a structured governance briefing on the activity beneath them. Not a notification feed. A digest covering what was decided, what Sentinel flagged, and where the knowledge picture shifted overnight.
A team lead sees their team. A practice head sees every engagement beneath them. The CEO sees the firm. Every briefing is scoped to what that leader governs, and nothing beyond it.
Activity
What moved in your organisation yesterday.
New decisions proposed and confirmed, canon updated, gaps identified and closed. The knowledge picture at a glance, across every scope you own.
Signals
Alerts that need your attention.
Policy drift, risk accumulation, coverage gaps. The signals Sentinel raises, distilled into what you need to act on before the day starts.
Scope-aware
Every leader sees what they govern.
The same governance boundaries that govern the system govern the briefing. Nothing surfaces above the scope it belongs to.
The signals don't wait for your stand-up. They're in your inbox before the day has started.
Built for governance
Designed for the people accountable for what gets said.
Every answer carries provenance.
Each response cites the decisions it draws on. Each confirmed decision carries the person who confirmed it. When the AI draws on a proposed decision, it says so.
Scope-bounded, governance-aware.
Content stays where it lives: marketing cannot read finance canon. But your governance hierarchy is real - parent-scope decisions bind their children, and cross-team patterns surface to the level that owns them. Leadership sees signals, not raw data. Cross-scope sharing is explicit and audited, never accidental.
Bring your own model.
Anthropic, OpenAI, Google Gemini, or an OpenAI-compatible endpoint, including self-hosted. Constans is the governance layer; the underlying intelligence stays your choice.
Your content stays in your tenancy.
Constans does not train on your content, and the LLM providers we ship with (Anthropic, OpenAI, and Google Gemini paid APIs) do not either, by their default terms. DPA available on request.
Audit trail.
Reads of canon, every confirmation, and every change to decisions and canon are recorded with timestamp and actor. Exportable for compliance review.
AI you can stand behind.
When your AI answers a customer, it answers from your sanctioned knowledge, not from whatever the underlying model invented. When an auditor asks how a decision was made, the trail is in the system.
Governed knowledge in every tool your team uses.
Constans ships as an MCP server. Your canon is available inside Claude Code, Cursor, Windsurf, and any MCP-compatible AI agent. When your team's tools reason about your organisation, they draw from governed knowledge, not the model's best guess.
Why not just a wiki?
Every tool you already have stores knowledge.
None of them govern it.
The shared wiki
It can't tell you what's still true.
Notion, Confluence, a shared drive - they hold documents, but the current decision and last year's abandoned draft sit side by side, equally confident. Knowledge goes stale silently, and the space quietly becomes a graveyard nobody trusts.
The AI assistant
It answers from whatever it finds.
A general chatbot or a bolt-on search retrieves fast and sounds certain - but it draws on whatever it can reach, sanctioned or not, current or not, with no provenance and no human in the loop. Confidence without governance is exposure, not an answer.
The people who know
It walks out of the building.
Asking the person who decided works right up until they're in a meeting, on leave, or gone. The reasoning lives in a head, not a system. When it leaves, so does the answer - and you're debating it from scratch.
All of them hold knowledge. None of them govern it - human-confirmed truth, provenance on every answer, the gap captured when something is missing, drift caught before it costs. Once a decision has to be answered for, storing it was never the hard part.
For consultancies and advisory partners
Your engagements end in slide decks.
Constans is where they live instead - as an AI expert your client keeps.
Run your own firm on Constans first - your methods, your decisions, your judgement, governed and answerable. Then bring it to the clients you advise.
Run your firm on it. Then bring it to your clients.
The work you do for a client currently lands as a deck that gets filed and forgotten. With Constans it lands as living knowledge they keep - an AI expert tuned to every decision and conversation, that goes on paying back long after the engagement ends.
Your IP. Their org. Lasting value, captured.
Assurance
Your knowledge stays yours.
Bring your own model.
Anthropic, OpenAI, Google Gemini, or a self-hosted endpoint - your key, your provider relationship, your terms. Constans is the governance layer in between, and never trains on your content.
Traceable and isolated.
Every answer cites the decisions it drew on, every confirmation and change is on the audit trail, and each tenant is isolated at the infrastructure level. Nothing crosses a boundary it wasn't given.
Yours to leave with.
Full data export and tenant deletion, whenever you ask - no lock-in at the data layer. DPA available on request.
Introduce Constans into your organisation.
A 30-minute conversation is the right starting point. We'll talk about your governance setup, where decisions currently get lost, and whether Constans fits.