Mycroft

The on-device investigator harness.

Mycroft runs in Osaurus on your Mac. Plan research with Navigator, investigate with Spotlight, keep watch with Scoutpost, and connect what you find in your OpenKnowledge wiki.

Website · Install with Indicator Lab · Manual setup · Contributing

Install

Indicator Lab provides guided setup, connections to the tools, repair and updates. Mycroft's Osaurus setup is for Mac; the installer checks compatibility before installation. Managed installation is included for members of Indicator’s Lab tier.

Buried Signals Engine (bsig) handles installation for both Indicator Lab and manual users. It installs the versions pinned in its signed catalogue. Cloning this repository previews the website and gives you the source; it does not configure the agents or install their models.

For manual or agent-led installation, follow signed Engine setup. The old install.sh deliberately exits without installing anything.

For an existing Engine installation, bsig configure plan mycroft prepares an install plan and bsig keys list identifies optional credentials. See the manual guide for configuration input and private key entry.

Agents

Agent	Role	Where it runs
Mycroft	Searches your wiki, finds OSINT tools with Navigator, works with Scoutpost monitors and extracts fields from documents.	On-device in Osaurus
Spotlight	Runs an investigation, with your review of the method and findings.	On-device in Osaurus
Analyst	Connects findings and proposes sourced additions to your wiki. You confirm writes before they land.	On-device; available through Osaurus and OpenKnowledge's Agents panel
Watson	Does web research when you select the cloud option. It has no wiki access.	Cloud, through OpenRouter using Mistral EU endpoints with zero data retention

On-device inference does not make every connected service local. Navigator and Scoutpost are hosted services. Osaurus asks before calls that create something or send text to them. Keep confidential source identities out of external requests.

The system instructions are in instructions/. The Analyst's Pi and ACP dependencies are pinned in analyst/; Engine installs them beside the managed checkout. Model choices come from Engine's signed catalogue rather than this README.

Skills

The installation and the website catalogue serve different purposes.

skills.manifest lists six installed skills, each exposed as a slash command on Mycroft:

Skill	Purpose
navigator	Find OSINT tools and methods
scoutpost	Set up monitors and read their findings
foia-requests	Draft and track public-records requests
interview-prep	Prepare questions, background and attribution rules
story-pitch	Frame a story pitch
ai-writing-detox	Remove formulaic AI writing patterns

The website features four skills and methods: fact-checking, public-records requests, interview preparation and story pitches. It omits the tool-connection skills and AI-writing cleanup. Fact-checking is an archived method reference, credited to Verso and Buried Signals; it is not included in the current Osaurus installation. Its source and status are recorded in catalog/featured-skills.json.

The four installed journalism skills are adapted from Joe Amditis's MIT-licensed work. Each retains its credit; see third-party notices and the provenance record.

Repository guide

Path	Contents
index.html, DESIGN.md	Static landing page and its visual rules
assets/, fonts/	Product media, partner marks and typefaces
instructions/, skills/	Agent instructions and installed skills
analyst/	Pinned Analyst runtime dependencies
install/contracts/	Install schemas, templates and fixtures imported by Engine
catalog/	Signed Engine catalogue mirror and website-only featured methods
runtimes/	Runtime-pack release record and licence notices
tests/	Website, catalogue and installer-pointer checks

Preview the site with python3 -m http.server 8765, then open http://127.0.0.1:8765/. There is no site build step. See Contributing for checks and deployment.

Credits and licence

Built by Buried Signals, with Osaurus and OpenKnowledge. The Analyst uses Pi; the website uses Three.js. Model versions and availability are controlled by Engine. Media provenance is recorded beside the preview clips and screenshots.

MIT · Third-party notices · Security