- Python 100%
| agents | ||
| benchmarks | ||
| contrib | ||
| docs | ||
| examples | ||
| methods | ||
| reports | ||
| skills | ||
| tests | ||
| tools | ||
| AGENTS.md | ||
| CONTRIBUTING.md | ||
| LICENSE | ||
| README.md | ||
| ROADMAP.md | ||
| TODO.md | ||
truth-kit
truth-kit is a Kyvero Vexus project for building practical infrastructure for
truth-seeking research.
This repo focuses on three things:
- finding and recovering information (including hard-to-find and archived material)
- tracing information lineage (who sourced what from where)
- keeping reasoning disciplined under uncertainty
LLMs are part of the toolkit, not the authority. The project treats model output as assistive and requires evidence trails, provenance tracking, and reproducible workflows.
What this repository builds
truth-kit is intentionally a mixed-format toolkit, not just a prompt collection.
Expected components include:
- method docs and research protocols
- CLI tools and reusable libraries
- workflow/skill wrappers over those tools
- benchmark datasets and scoring tasks
- reports and worked investigations
- agent interfaces that orchestrate the above
Scope map
Current workstreams (summarized):
- discovery and retrieval
- archives and recovery
- reverse search
- provenance graphs and lineage tracing
- source criticism and textual lineage
- media/authorship forensics
- research protocol design
- benchmark and evaluation systems
- live trend lurk and early-signal monitoring
For detailed scope, priorities, and phases, see ROADMAP.md.
Architecture direction
Long-term UX goal: a simple truth-seeking agent (or agent system).
Implementation strategy: build bottom-up and keep the core inspectable.
Layering model:
- methods
- tools
- workflows/skills
- agent interface
- multi-agent orchestration only if benchmarks show clear gains
Architecture details are in docs/architecture.md.
Design stance
The project prioritizes:
- evidence quality over rhetorical polish
- provenance visibility over black-box conclusions
- calibrated uncertainty over fake certainty
- reproducibility over one-off demos
Canonical principles live in docs/design-principles.md.
Reporting expectations live in docs/reporting-format.md.
Current status
The repository is in early design + scaffold mode.
Completed so far:
- scope and roadmap
- architecture document
- contribution guidance (human + agent)
- AGPL licensing
- initial directory skeleton aligned with architecture
- baseline design-principles and reporting-format docs
- dork capability mapping + migration notes
- shared retrieval/provenance schema draft + archive/reverse contracts
- first 3 dork-derived benchmark fixtures
- DorXNG fork established under KVC (
Kyvero-Vexus/DorXNG) - lurker lane design + trend event/alert schemas for tooling-first live monitoring
- lurker phase-2 planning artifacts (RSS adapter contract, baseline source profile, scoring defaults, TrendAlert adapter spec, trend benchmark fixtures)
Read next
- ROADMAP.md
- TODO.md
- docs/architecture.md
- docs/design-principles.md
- docs/reporting-format.md
- docs/dork-capability-map.md
- docs/dork-migration-notes.md
- docs/lurker-lane.md
- docs/mvp-use-cases-and-issue-specs.md
- docs/mvp-test-cases.md
- tools/common/SCHEMA.md
- tools/common/RSS-ADAPTER-CONTRACT.md
- tools/common/TREND-SCORING.md
- tools/common/TRENDALERT-INVESTIGATION-ADAPTER.md
- tools/common/sources/lurker-baseline-sources.json
- tools/archive/CONTRACT.md
- tools/reverse-search/CONTRACT.md
- CONTRIBUTING.md
- contrib/README.md
- AGENTS.md
License
GNU Affero General Public License, version 3 or later. See LICENSE.