One platform under every process. Built with you, in days.
Nine years on one platform. Each new application starts from the connectors, the permissions, the agents’ harness and the channels the last one left behind, so a department’s process becomes a working application in days, on top of the systems you already run.
How an application is built
A conversation, then a working screen.
No eighteen-month project and nothing to replace. You bring one process; you see it running on a sample of your own data before you sign anything.
Where the agents run
The agent carries the step. A person decides.
Agents run inside the application your team works in, with the same permissions as everything else. AI proposes; people approve. You widen what an agent may do on its own as trust builds.
What an agent may do
- Read what its permissions open, and nothing else
- Rank the day’s work and say why
- Research, draft and answer from the live data
- Propose an action, as a row a person takes or sets aside
What a person keeps
- Anything that leaves the building, such as a message to a customer
- Sensitive actions, with explicit sign-off
- How far each agent’s autonomy goes, set by you
- The decisions the law or the customer depends on
Your data, your systems
Read where it lives. Nothing replaced to start.
The application connects to what you already run and keeps its own records in a database of its own. Your systems stay the systems of record for what they own.
Your ERP, your point of sale, your maintenance system, an AS/400. Data is read where it lives; nothing is replaced to start.
44 adapters · 352 data movementsOne database and one isolated worker per application. Pages, rules and prompts in git. Export at any time. If we are ever gone, you keep running it.
the continuity licenseClaude by default, on our keys at cost or on yours, never trained on your data. When a better model ships, the application takes it.
no third-party trackersSeveral agents, several users, each person with their own permissions on every read and every write.
roles and row-level securityThe web, native iOS and Android apps, SMS, WhatsApp, email, Teams and voice. One application, one set of permissions across all of them.
web · iOS · Android · voiceTools and limits per agent, an audit trail and a replay of every turn, cost caps, and a person’s approval on anything that leaves the building.
every action logged with its reasonIndependently reviewed
Security and governance
As our security page states it.
Each line below is the current public statement, shortened and never strengthened. The full pages stay where they are.
Amazon Web Services, us-west-2 (Oregon), with the backup replica in us-east-1. Each application has its own PostgreSQL database, storage paths, worker microVM and token signing key.From /security, last updated October 17, 2026.
Requests enter only through Amazon CloudFront and AWS WAF; the application service, databases and microVMs are not reachable from the internet directly.From /security, last updated October 17, 2026.
At rest under AWS KMS or S3 server-side encryption (AES-256); in transit over TLS; secrets encrypted with AES-256-GCM under KMS-wrapped keys, never written in code or logs.From /security, last updated October 17, 2026.
Staff reach production under individual identities with multi-factor authentication, through audited doors; access by role, least privilege, reviewed every quarter.From /security, last updated October 17, 2026.
Every component logs centrally; production logs are kept 365 days and never carry a secret. Every error-level event becomes a tracked issue, closed with its fix.From /security, last updated October 17, 2026.
Nightly backups of each app’s database and files; seven days of point-in-time recovery; database backups replicated to us-east-1. Recovery objectives: 1 hour / 4 hours for lost data, 24 hours / 8 hours for the loss of the region.From /security, last updated October 17, 2026.
Every incident is rated and handled within its severity’s target. Affected customers are notified within 72 hours of confirming an incident that affects their data.From /security, last updated October 17, 2026.
Controls mapped to the SOC 2 Trust Services Criteria for Security, Availability and Confidentiality; the controls matrix on request. A CASA Tier 2 Statement of Validation from TAC Security, issued August 7, 2026, valid until August 8, 2027. Twenty policies under NDA.From /security, last updated October 17, 2026.
AI proposes. Humans approve. Sensitive actions require explicit sign-off. Every action is logged and auditable: what agents did, why, and when.From /governance.
People and agents see only what they are allowed to see and do only what they are allowed to do, with one set of permissions across every surface.From /governance.
Book a walkthrough
Bring one process.
Leave with a working screen.
Bring your aging report, your worst ticket or the spreadsheet everybody is afraid of. The walkthrough shows the desk on a sample of it on the same call.