Tools

Tools, demos and research

Public tools, synthetic demos and private research have different limits. Each entry states its availability, evidence and open questions. Working tools stay on their own sites; these pages explain them.

BetaPrecision not yet measured

RADAR

Watches AI platform changes and security advisories, and publishes what changed, why it matters and what to do.

Agent network boundariesRADAR’s report on Claude Managed Agents and allowed-host enforcement.Read ↗ (opens in a new tab)
Agent import risksRADAR’s report on code injection and external-reference handling in the AWS Bedrock agent toolkit.Read ↗ (opens in a new tab)
Developer platform securityRADAR’s report on a TechDocs configuration vulnerability in Backstage.Read ↗ (opens in a new tab)

Selected items · checked 8 Oct 2026

Selected examples, not a complete feed or a measured accuracy sample. Model confidence is not measured correctness.

How RADAR works Open RADAR

Public alphaFictional cases

CCP Study Lab

Practice evidence review through fictional cases and original questions, in your browser.

AI-authored alpha content · independent review pending

What it does

Start from a confident security claim made by a fictional company, inspect the supplied record, then connect the evidence to a reasoned conclusion.
Local-first practice in your browser, with no account or uploads.
Public v0.8.3 Research Alpha, observed 15 Oct 2026. Practice scores are not credentials or certification.

Explore CCP Study Lab

PrototypeSynthetic demo

PROBE

Adversarial assessment research for AI systems, with a public workbench you can step through.

Fictional scenarios, fixed responses · no live assessment yet

What it does

Walks through attack scenarios one step at a time: the input, the system’s response and the finding.
The report it is being built toward maps findings to the OWASP GenAI and Agentic Top 10.
The synthetic workbench is open at probe.trulysovereign.com. The demo does not score your system or run live AI.

Explore PROBE

In active developmentHosted runtime · private access

JCR

A checkpoint for AI agent actions, in active development, with a receipt for every decision.

Hosted runtime, private access · the public demo is synthetic

What it does

Actions routed through it are checked against policy, then allowed, denied or held for a person to approve.
MCP gateway: an agent that uses the Model Context Protocol can be pointed at JCR without changing the agent.
Evidence packs: an export of receipts and the policies they cite, designed to be checked offline.

Explore JCR

ResearchNothing public to check yet

FARADAY

Explores running AI models and embeddings on hardware you control.

Nothing public to check yet · egress measurement intended

What it is working toward

For environments where the data cannot leave, and teams that have restricted public AI and need a sanctioned place to send people instead.
The intended public proof is an egress measurement with a stated configuration and scope.
No production deployment, benchmark or date. Private test results are not public product acceptance.

Explore FARADAY

ExperimentUnproven

Undertow

An experiment in hearing a shift in public language before it becomes the headline.

Personal research tool · forward-test start unverified

What it does

Designed to look for shifts in public writing: new phrases, changes in tone, and the same idea appearing in places that don’t usually agree.
October 12, 2026 was the target for first unscored alerts; the start is unverified here. October 19 remains the target for scored alerts.
Runs privately, with a walkthrough on request. Unproven: no forward-test results are published here.

Explore Undertow

Additional public interfaces

Available to inspect, with explicit limits.

CITECHECK

Early access, in development. Citation existence checks are available; semantic support checking and its report are not ready.

Individual tool page →Open CITECHECK ↗

LEDGER

In development. The public start screen offers a register, a fictional demo and JSON import. Interface observation does not establish runtime correctness or production acceptance.

Individual tool page →Inspect the public interface ↗Ask about LEDGER →

The December build target retained in the roadmap below is historical; it does not describe current public availability.

Roadmap

Development notes and plans

CITECHECK and LEDGER have public interfaces; their current scope is described above. These notes retain earlier targets and unfinished goals. MANIFEST remains planned. Dates are targets, not completed milestones.

Illustration · support check coming, not live

Early accessIn development

CITECHECK · development notes

Finds the citations in AI-written compliance documents and checks that each one exists in its source. Checking whether they say what they're cited for is coming.

Early access from Oct 9, 2026 · build target Nov 8, 2026. Build target from the earlier plan; see current scope above.

Live now, and what’s coming

Live now: finds the regulatory and framework citations in a document, grouped by source, in your browser, including NIST SP 800-171 and CMMC Level 2.
Live now: checks each one exists in a version-pinned copy of its source and shows the clause text where it is found. Coming, not live: whether that text supports the claim, and a report.
Coming, not live: will publish its own measured error rate from a hand-labelled test set.

Explore CITECHECK

Illustration

In developmentPublic interface

LEDGER · earlier design goals

An AI use register and snapshot kit for companies with no AI governance owner.

Earlier build target: Build about Nov 16 – Dec 13, 2026. Public interface now available; see notes above.

Earlier design goals · not validated here

Lists every AI use, including AI features inside the SaaS you already pay for, with an owner and evidence.
Designed to run entirely in your browser, so the register stays on your machine.
Produces a short Snapshot: what you use, who owns it, and what's missing.

Explore LEDGER

Illustration

Planned · RADAR channel, pilot early 2027

MANIFEST

A scorecard of how SaaS vendors use customer data for AI.

Pilot about Dec 14, 2026 – Jan 3, 2027

What it will do

Ships as a RADAR channel, so it reuses RADAR's change tracking and public corrections.
Tracks each vendor's AI and data-use terms and flags when they change.
Pilot of 10 vendors, each score human-reviewed with links to the evidence; starts after RADAR's precision sample is complete.

Get notified

Status, plainly

Every tool carries one of nine labels. A prototype stays a prototype until it has evidence.

What these labels mean

LiveA running public service you can use today. No guarantee of accuracy or uptime.
BetaA running public service, still being tuned. Expect changes. No guarantee of accuracy or uptime.
Public alphaPublic and usable, still changing. Independent review pending.
Early accessLive and working on your own input, but unfinished. Features are still being added.
PrototypeA working demo on synthetic data. Not a product.
In developmentA public interface is available to inspect. Feature-level correctness and production readiness are not established.
In active developmentUnder active development. Runtime access is private; the public demo is synthetic.
ResearchBeing built. Nothing public to use yet.
ExperimentA personal research tool. Unproven; access and test status are stated on its page.

Roadmap

In designScoped and specified; no code yet.
PlannedOn the schedule; not started.

We publish where we were wrong.

RADAR publishes its item record and corrections, including edition corrections with their own scope. Those records are useful evidence, but they are not a measured error rate. Random review progress and measured precision are shown separately on RADAR’s page.

RADAR’s live record →Item and edition corrections How precision is counted