Question 1

What is Kentron AI and what does it actually do?

Accepted Answer

Kentron AI is an autonomous AI coworker platform that gives your team an AI agent with real execution capability: it runs code in a private isolated sandbox, reaches your connected tools (AWS, Jira, GitHub, Vercel, Slack, and more) through scoped actions, and converts probabilistic AI requests into deterministic, auditable outcomes. Every action is logged as an immutable, cryptographically chained “receipt”—the permanent source of truth—enabling full audit, replay, and crash recovery.

Question 2

How is Kentron AI different from Claude Code, Codex, or other AI coding tools?

Accepted Answer

Claude Code and Codex are single-session coding assistants for individual developers; Kentron AI is a multi-tenant production control plane that can orchestrate and run AI agents like those as auditable background workers. Kentron AI adds durable execution (survives crashes), deterministic replay (proves what happened), compliance logging (immutable audit trail), multi-tenant isolation (each org’s data is separate), and integration with the systems you connect—all things you need for production governance that a session-based coding tool doesn’t provide.

Question 3

Is it actually safe to give AI agents access to my production systems?

Accepted Answer

Yes—if you use Kentron AI correctly. Every action runs in an isolated OpenSandbox with zero ambient access; credentials are never stored in Kentron AI (Nango owns them, Kentron AI only stores encrypted references); all state-changing operations are deterministic, replayable, and cryptographically chained so you can verify exactly what happened; and Kentron AI enforces org-level policy gates before any AI decision is executed. The default is “sandbox first, audit everything”—not “trust and react.”

Question 4

What’s the difference between a "receipt" and regular logs or observability?

Accepted Answer

Observability logs describe state that exists elsewhere; deleting logs loses visibility but not data. Receipts are the state—the append-only source of truth—and the running system is just a derived copy. Deleting receipts loses everything. This matters because it means Kentron AI doesn’t copy data around; it owns the provable history. You can replay any past moment, audit any decision, and prove causation—not just observe failures after they happen.

Question 5

What SaaS apps and systems can Kentron AI connect to?

Accepted Answer

Kentron AI carries a catalog of 1,000+ connectors (AWS, Google Cloud, Vercel, GitHub, GitLab, Jira, Confluence, Slack, HubSpot, Salesforce, Linear, Zendesk, and many more). You control which integrations each team member can access via org-level policy; credentials stay with the provider (Nango), never in Kentron AI; and you can add new integrations without waiting for a Kentron AI release. If Nango supports it and your team has a connection, the AI can use it—sandbox first, audit everything.

Question 6

How do I know the AI actually did what it claimed to do?

Accepted Answer

Every action is a receipt: immutable, timestamped, cryptographically chained to the previous receipt, and contains the command, the evidence artifact, and the exit code. You can replay that receipt independently to verify the outcome. For external APIs (AWS, Jira, GitHub), the evidence includes the raw API response verbatim plus a tamper-evident hash. For code execution, you get the stdout, exit code, and any files created. Replay + independent verification = proof. No copying data between systems, no hidden side effects.

Question 7

What happens if the agent makes a mistake, gets stuck, or a task crashes?

Accepted Answer

Kentron AI logs every step as an immutable receipt, so if a task crashes, you can replay from the last receipt without re-running earlier work—crash recovery is automatic. If the AI makes a wrong decision, that’s logged and auditable; you can review the evidence, fix the underlying issue, and replay from any point. If the agent gets stuck, Kentron AI detects stalled work (expired leases, watchdog, invariant checks) and self-heals with no human intervention—re-folding receipts and re-dispatching deterministically.