Two ladders. Both honest.
One ladder says how much freedom an agent has earned. The other says how finished each thing we sell really is. You can read both for everything on this site.
How much may an agent do alone?
Four levels, set per kind of work. An agent can draft customer replies for a person to approve at L1 while it files internal notes at L2. Select a level.
L3 Decide. The agent makes one agreed kind of low-risk decision inside its limits and records why. A person reviews the log and can reverse it. Measured by: Decisions made, how often people reverse them, and hours saved.
- A class of work becomes eligible for the next level after 50 items with 95% used without an edit.
- Customer-facing work has a stricter test, and a message to a customer always waits for a named person's approval.
- Eligible is not promoted. A named person decides every step up.
- Two corrections inside 14 days and the capability steps back one level until it is retested.

Real product screen on fictional Example Co data.
What is live today.
We label everything. Live means in daily use in our own operations. It does not mean installed for you yet: every client gets a dedicated install, configured for their roles.
In daily use in our own company. The numbers on this site come from these.
Running with limited scope: a few mailboxes, one team or one kind of work.
Built or designed, and not yet in daily use.
| Name | Kind | Maturity | What it is |
|---|---|---|---|
| Assay | App | Live | An auditor kept apart from the builders, so nobody marks their own work done. |
| CAD Nexus | App | Live | Our complete electrical and mechanical CAD engine. People and agents design, test and build control panels in it. |
| Case Nexus | App | Live | Our legal product. A case workspace that assembles the evidence, the documents and a sourced timeline for a legal matter. |
| Cluster | App | Live | An inventory of every agent, service and app, with its owner and its health. |
| Connectors | App | Live | The systems your agents read and work in. |
| CrystalAgent Core | App | Live | The runtime every agent is built on. One definition per agent, used on every channel. |
| Facet | App | Live | Cited research on demand, with the cost of each answer shown. |
| Lattice | App | Live | One shared company memory. Every agent checks it before acting and adds to it after. |
| Loupe | App | Live | One tile per agent: what it is doing now, who owns it, and what is waiting on a person. |
| Loupe Crystal | App | Live | Eight measured corners per agent in one shape you can read at a glance. |
| Loupe Mind | App | Live | A live stream of every step an agent takes, with the reasoning attached. |
| NexTalk | App | Live | A live voice room where you talk to your agents and they talk to each other. |
| Quarry | App | Live | Quotes, projects, work orders and accounts in one system your people and your agents both work in. |
| Ashley | Agent | Live | Executive Assistant |
| Audrey | Agent | Live | Compliance and Supplier Operations |
| Claire | Agent | Live | Customer Desk |
| Dot | Agent | Live | Finance and Invoicing |
| Iris | Agent | Live | Research |
| Karen | Agent | Live | Program Management |
| Miles | Agent | Live | Audit Dispatcher |
| Reese | Agent | Live | Sales and Revenue Risk |
| Rex | Agent | Live | Chief Software Architect |
| Simon | Agent | Live | Orchestrator |
| Vera | Agent | Live | Independent Auditor |
| Loupe Economics | App | Pilot | An agent profit and loss, built from measured outcomes. |
| Loupe Oversight | App | Pilot | Everything waiting on a named person, and what each agent may do alone. |
| Prism | App | Pilot | Shared-inbox triage that tags, numbers and routes every request. |
| Nora | Agent | Staged | Travel Desk |
| Owen | Agent | Staged | Field Engineering |
Maturity as of November 2026. We update this page when a label changes.
Start at level zero.
Stage 0 is read-only. We watch how the work gets done and nothing in your systems changes.

