One place to install and maintain the tool suite

WINWindows

MACmacOS

LNXLinux

Five tools for the full investigation cycle

MycroftRun the suite through Goose with a local model, AI subscription, or zero-data-retention provider.

SpotlightMethodology, OSINT research cycles, evidence, fact-checking, and reusable case archives.

ScoutpostMonitor pages, social profiles, councils, aircraft, vessels, and your beat for leads.

NavigatorFind OSINT methods across Indicator’s archives and more than 13,000 tools.

SplashTurn sources and data into reviewed charts, maps, stories, and visuals.

Use each tool independently—or connect the full suite through Mycroft.

Practical help when you need it

Learn from working examples and get direct help with your own setup.

Four ways these workshops can earn the hour

• Investigation debriefs: journalists unpack a real workflow—including where AI helped and where judgment stayed human.
• Newsroom operating tours: how teams monitor sources, surface leads, and run recurring reporting work.
• Practical workshops: build or test a method together, with the setup and failure points visible.
• Deep-dive workshops: start with the basics, then go deeper—like this Going Sovereign workshop.

Build reporting workflows you own

• Keep sensitive material on infrastructure you control.
• Turn recurring reporting work into a method colleagues can run again.
• Keep the model replaceable: use cloud for open material, local when custody matters.

A model is a very good next-token guesser

• It reads the context you give it and predicts what should come next.
• It cannot fetch evidence, write a case file, or wait for an editor on its own.
• Without a harness, the only thing it can return is more text.

The harness is everything around the model

A GUI around a harness around a model

That is the useful shorthand for Claude or ChatGPT Desktop. The harness supplies context, tools, and state; the model does the reasoning.

You do not have to build a harness from scratch

Spotlight is an agent system for running OSINT investigations

It takes a lead through methodology, research cycles, evidence archiving, independent fact-checking, and a reusable case archive.

Making Spotlight local produced a transferable blueprint

The same pattern can create custom sovereign workflows for any newsroom or company—run through its existing harness, or through a small local model when needed.

Add validation or human approval only where the workflow needs it.

Example: turn a council agenda into a lead brief

The same controlled path runs next week. The journalist still decides whether there is a story.

One blueprint can support very different workflows

How do you move a frontier workflow onto a small model?

Spotlight already worked with frontier models. The local 12B failed because the frontier-shaped harness overwhelmed it.

That engineering path—not a model-size comparison—is the Going Sovereign experiment.

The small-model path became usable

Before and after use the same 12B and laptop. The difference is the local-engineered harness plus the narrow procedure adapter.

A small helper model reads the messy stuff first

RLM stands for Recursive Language Model: in Spotlight, it is a second, smaller model with one job—compress large, noisy inputs before the main investigator sees them.

Use the pattern for: meeting transcripts, filings, long PDFs, monitoring feeds, or any input too noisy to send directly to the main model.

The sovereign topology

Use the largest model your infrastructure comfortably supports

Do not fine-tune to save a working model. Tune only when hardware or custody constraints make the smaller model the useful path.

How do you teach a small model the procedure?

Spotlight used 14 real trajectories and 12 targeted corrections. Total RunPod spend across the iterations was about $10.

How can you use something like this?

Add validation or human approval where mistakes would matter. They are controls you choose—not the point of every workflow.