How Inficy works

You connect an agent, its executions appear, and you can read what happened. Everything after that, preserving, exporting and optionally certifying a record, builds on the same captured activity.

1. Connect an agent

You copy one command from your workspace and paste it into your agent project. The installer detects the runtime, exchanges a one-time setup code for a reporting token, stores it as a secret and never prints it, then runs a connection self-check. Setup codes expire in ten minutes and can be redeemed once.

2. Executions appear

Each thing the agent does end to end is recorded as an execution made of ordered actions: tool calls, model calls, web requests, file writes, handoffs, human approvals, failures and recoveries. Every event carries an idempotency key, so retries never create duplicate history, and ingestion is queued so the agent stays fast under load.

3. Understand the activity

Inficy groups executions so you can see where agent activity actually goes: which tools and services were used, which domains were reached, how much wall-clock time passed against how much of it was active work, where a human was waited on, and, where observable, what the agent tried to spend. Everything shown is derived from captured execution activity, so an action nobody reported is not inferred.

4. Inspect one execution

Open a single execution to read it step by step with timestamps, durations, tool and model identifiers and outcomes. Failures and interruptions stay in the timeline where they happened rather than being filtered out, and payment states are shown exactly as observed: requested, attempted, approved, succeeded, failed or refunded. A request is never displayed as a completed outcome.

5. Preserve, export and verify

Executions are stored as canonical NexArt artifacts, so a record is tamper-evident rather than a mutable log line. You can export an execution and its evidence, and anyone holding the export can verify it offline.

6. Optionally certify a selected execution

Certification is deliberate, per execution and paid in credits. Inficy submits the artifact to the NexArt Node, which attests the exact bundle and returns an independently verifiable certificate. Certification is idempotent, so the same execution cannot be double-charged. Most executions are never certified, and nothing about recording or inspection depends on it.

Why the record is trustworthy

Executions are recorded as canonical NexArt artifacts and verified offline with the NexArt SDK before they are trusted in the timeline. An artifact that fails verification is shown as unverified rather than silently accepted. Verification and certification are different things: verification checks that the artifact is internally consistent and unaltered, certification adds an independent third-party attestation of that same artifact at a point in time. A certified record shows an execution existed in exactly that form; it does not show the agent's decision was correct or safe.

Confidential executions

When an execution contains sensitive content, it is sealed: the artifact commits to the content without disclosing it, and the opening material stays with the workspace owner. Verification still works; the payload stays private.

Continue

See the documentation for the connection and API details, or the security model.