Sample data
Chain
POSTURE · READING
Welcome back.
Reading the ledger.
Connect your first AI application
Nothing is reporting to this workspace yet. Every page here fills from what your applications send: each call an agent makes, what it touched, and what was stopped. Three steps, about ten minutes.
- Issue a credential for the application One per application, shown once. It lets that application send records to this workspace and do nothing else.
- Connect it
Choose how your agents reach their tools. Put the credential where it
says exc_….
- Run your agent once
The first record appears here within seconds, and this panel goes away.
Waiting for the first record. This page checks every ten seconds.
This workspace has no platform agreement in force. Its evidence and Code Review
stay readable; the rest of the console, and the assistant, return with an
agreement. See what a console covers.
The ledger and its reports are open to this workspace's Security Officer and its
backup only, and you hold neither role. Code Review and Settings stay open. A
workspace administrator can ask for the role to be assigned in Settings, and a
second person approves it.
Answers come from this workspace's ledger, and every question and answer is recorded. Your conversations are kept in Spaces, under Chat history, for Security Officers of this workspace.
Needs attention Anything waiting on a person
Current state What the ledger says right now
Modules
Open a module to see what it found
Spaces: case files, your conversations, and what you chose to keep
Chat history arrives on its own: each question you ask Excubitor and the answer you were shown, with the date and time. Briefings arrive on their own too, written from the ledger after each day and each week. Everything else is here because someone kept it, and keeping or removing an item is recorded in the ledger like any other action. A kept item is a copy on a shelf, not evidence. Cases gather records, items, periods, smart folders and notes for one incident, agent, audit or quarter, and export as one proof bundle.
Settings
Your agreement, the applications reporting here, the people who can see them and how they sign in. Issuing a credential and adding a person are administrator actions, and both are recorded in the ledger with your name on them.
Nothing here is yours to change
Issuing credentials and managing people belong to an administrator of
this workspace. Ask one of them, or the person who set the workspace up.
Events
Every record in this workspace's ledger, newest first. Open one to see what it holds.
Help & support
How this console is organised, what each page's figure means, and what to do when one needs you. Every figure comes from the signed ledger your applications report to.
Customers
Start an evaluation, set up a paying customer, change terms, resend an activation link or download a licence file. Evaluations start as soon as you confirm them. Anything a customer pays for waits for a second person to approve it. Everything here is recorded with your name.
Activity
Every change to this workspace and every decision taken in this console, with the person who made it. Administration comes from the workspace audit log; decisions come from the signed ledger and cannot be edited after the fact.
Decision Replay
Each step is one thing an AI did, oldest at the top. Read down to see how a simple request became a $48,200 payment. Click any step to see its signed record.
⚠
This $48,200 payment was poisoned 3 steps back. Hidden instructions in a document, not your request, created it. Excubitor traced it to the exact source and held it for your approval.
Contagion: how a problem spread
Whether a flagged input reached more than one agent, and the path it took.
Behavior: who acted out of character
Excubitor quietly learns what "normal" looks like for each AI helper. When one suddenly does something it has never done, that stands out, a strong sign it was tricked or hijacked.
Guard: every call, wrapped
Guard sits around each model and tool call: it checks the arguments, holds the rate limit, and records what happened. This is that record, gathered by tool.
Detect: what the detectors caught
Injection, PII and secret detectors run over the payloads passing through. Everything they raised is here, with what it maps to.
Gateway: in front of the tool servers
Every call an agent makes to another service goes through one chokepoint, so it is identified, rate-limited and recorded on its way out rather than noticed afterwards. This is what left the building.
Red Team: probes on your schedule
Adversarial probes, the standard set plus any your team or a Shield assessment added, run against a staging agent on a schedule you set. Each run is scored and signed into this ledger, so a defence that stopped working is noticed before an attacker finds it.
Assessment
A scenario engagement, so it carries no chain attestation and
cannot be mailed as a ledger report.
The full assessment, including the method notes
and every scenario summarised here, comes with the report.
Model providers
Every model call an agent made, hosted or on your own servers: which provider and model it went to, which model actually answered, whether it failed, how long it took and the tokens the provider reported. Built from the ledger, the same rows as the download.
Files taken in
Every file that reached an agent, from a model request, a tool, or your own parser: what it said it was, what its bytes are, its size and hash, and what was flagged. Never the content, never the file name.
Identities
Every person, workload or credential an agent acted for, and each service it reached. A settled identity that starts reaching new services is flagged, and several within the hour is flagged high.
Changes made
Every write an agent's tools sent to another system, and each change your application reported: the system, the kind of change, whether it took, and the tool call behind it. Ids are hashed, never stored.
Delegations
Every time one agent handed work to another: who handed it, who received it, how, the run it came from, the run it started, and who it was for. What was handed on is hashed, never stored.
AI register
Every agent this ledger has seen act, the applications it belongs to, what it did, and what was flagged, refused, held or contained. Download it for an insurer or an auditor: it names the ledger record it was computed through and carries a digest anyone can recompute from the ledger export.
Remediation
What was done about a problem, and by whose authority, as recorded on this ledger. A step marked automatic ran without asking; a step held for a person waits here for a decision.
Discovery log
Every finding, from every agent, across every repository: one signed, append-only log rather than a conversation per pull request.
Profile
One page per agent, application and tool, built from the ledger.
Excubitor by Aurakore · Every number on this page comes from your signed ledger. Times are Central time (America/Chicago).