What happens when it is wrong.
One gate that fails closed, and a record of every verdict.
One seam, and nothing gets past it
Whatever proposes holds nothing
Built and testedNo credential, no connector, no path out.
The gate is not ours to argue with
Built and testedSelah evaluates against your policies. OnDuty only executes.
Fail-closed by default
By designNo clean permit, no execution. Nothing configures that away.
Holds go to a person
Built and testedA held action waits. Nothing expires into being sent.
- 01Agentproposes · holds no credentials
- 02Decision enginepermit · hold · deny · your rules, not the model
- 03Executepermit → Connector
- 04A personhold
- 05Nothing happensdeny
Record: 0 rows
Each row carries the hash of the row before it. The real trail is kept and verified by the decision engine.
What isolation means today
| Concern | Today | On the enterprise tier |
|---|---|---|
| Tenant separation | Scoped by tenant. Other workspaces answer 404. | Same, plus a dedicated database on request. |
| Connector credentials | Encrypted at rest, key outside the database. | Same, with the key in your own key management. |
| Conversation content | Processed by a model vendor on our account. | Runners register today. Running agents there is not built. |
| Data residency | Not offered | Through Selah jurisdictions, scoped in the pilot. |
| Audit | Every decision in a hash-chained trail, verifiable. | Same, exportable. |
| Model training | Your data does not train shared models. | Same, in writing. |
| Third-party audit | No SOC 2 report | None yet |
What a pilot looks like
Six weeks, one workflow, one team.
Week 1
Scope one workflow, its allowed actions and the guardrails.
Needs your operation in the room.
A reply the example rule holds
- decision
- none yet
- shadow
- none yet
- the reply
- none yet
Week 1 · SCOPE · nothing is connected and no rule runs yet
In shadow a verdict is recorded, not enforced, so no tenant is called governed in production.
Three ways to start
Explore
No commitment: a non-binding letter of intent and first refusal on a pilot slot.
Pilot
Six weeks, one workflow, one team. Success criteria written before week 1.
Design partner
We build against your edge cases. Preferential pricing for 24 months.
Where we are looking for design partners
Where the founders have operated. There are no deployments yet.
| Sector | The shape of the mess |
|---|---|
| Vehicle retail | Long cycles, quiet quotes, history split across CRM, chat and inbox. |
| Hospitality | Heavy message volume, a small desk, one question asked three times. |
| Multi-location food service | Many locations; complaints look like a pattern only read together. |
| Venues and events | Huge demand, a short window, context right the first time. |
The questions that come up
What happens when the model is uncertain?
A confidence score is not a permission. Your policy permits, holds or denies, and a held action goes to a person with the full proposal.
Can it integrate with our stack?
Built: Google, Meta, HubSpot. Built, coming soon: Outlook, Slack, Salesforce, QuickBooks, Google Ads. Your own systems enter by REST, CSV, read-only Postgres or push. Some providers need app review first; the integrations page says which.
Can our own AI read it?
Yes, over MCP, on a key issued in your workspace. It can read and propose, nothing else. Signing in from Claude or ChatGPT is built, not yet exercised live. There is no developer API.
Who owns the data?
You do. It does not train shared models, and logs, history and the trail can be exported.
What does it cost?
Enterprise has no published price on purpose. Every other plan is priced on the pricing page.
How much of this is built?
The seam, executor, review queue, audit trail, Brain, connectors and billing run. Runners register but do not answer from your machine. No SOC 2 report; Google verification not submitted.