Trust & Operating Principles
Autonomous does not mean unaccountable. These are the rules every agent we operate is built under.
Guardrails live in code, not in prompts
Business rules that must never be broken — booking hours, slot conflicts, spend caps, contact policy — are enforced deterministically in the agent's action layer. A language model can interpret a request; it cannot override a hard guardrail. Every blocked action carries an explicit machine-readable reason.
Humans decide who gets contacted
Our lead qualification agent scores and routes — it is structurally incapable of outreach. Email drafts go through approval gates before anything is sent. Wherever an agent's output reaches a person outside your team, a human sign-off sits in between.
Every decision leaves an audit record
Each agent cycle writes an outcome record: what came in, what was decided, why, at what token cost. Bookings, rejections, scores, and handoffs are all attributable to a request ID and a loop cycle — so "why did the agent do that?" always has a concrete answer.
Your data stays scoped to your project
Client data lives inside the client's isolated project workspace with its own knowledge base, inbox, and loop memory. Agents are wired through swappable provider adapters — your systems are integrated per engagement, not pooled across clients. Nothing from one client's workspace trains another's agent.
Honest numbers: simulated vs. real, labeled
Every performance figure on this site that comes from our deterministic dry-run simulator is labeled as simulated evidence. Case-study result tables stay empty until a real pilot fills them with measured data. We do not publish estimates dressed up as results.
Self-improvement with a paper trail
Agents improve by writing structured lessons from their own cycles (a KB gap found, an error pattern caught). Lessons become recommendations to you — changes to guardrails or pricing-relevant scope are decisions made with you, not silently around you.
Escalation
Situations outside an agent's delegated authority — contract questions, unusual spend, potential security issues — escalate to a human immediately rather than being improvised. Boring and predictable is the point.
Questions about how an engagement would be scoped? Start here →