Skip to content

Git-Rocky-Stack/Agent-X

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

449 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent-X

Local-first AI document intelligence for Windows. Agent-X turns your personal document collection into a queryable, AI-augmented knowledge base — a native .NET 8 / WinUI 3 desktop app that runs entirely on your machine. Local by default: your documents, embeddings, conversations, and database never leave your device unless you explicitly opt in to a cloud model provider or web search.

License: MIT Platform .NET UI Tests

Latest release: v2.1.1 (installers below) · Current source: v2.1.2 — "Bedrock" security & supply-chain hardening; the signed v2.1.2 installer republish is pending code-signing certificate issuance. License: MIT — see LICENSE. Copyright (c) 2026 Rocky Elsalaymeh.


Download & Install

Installer Size Local model Best for
AgentX-Setup-x64.exe (SLIM) ~230 MB Downloaded on first run Most users
AgentX-Setup-x64-offline.exe (OFFLINE) — linked from the release notes ~2.1 GB Llama 3.2 3B pre-bundled Air-gapped / offline installs

Requirements: Windows 10 build 19041+ or Windows 11, x64. No account, no subscription, no telemetry.

What Agent-X Does

Knowledge Vault & ingestion

  • Import PDFs, Word documents, Markdown, text, code files, and web pages — single files, URLs, or entire folders
  • Automatic chunking, embedding, auto-tagging, and duplicate detection on import; collections, annotations, and full document management
  • Smart Inbox triage queue: captures from the browser extension, mobile companion, and plugins arrive with AI-drafted previews (suggested summary, collection, tags) for one-click accept / defer / reject
  • Audio transcription turns recordings into searchable documents

Search & retrieval (RAG)

  • Ask Your Files: natural-language questions answered from your own documents, with citations
  • Hybrid retrieval: HNSW vector search + SQLite FTS5 (BM25) keyword search fused with Reciprocal Rank Fusion
  • Retrieval pipeline with HyDE (hypothetical document embeddings) and LLM reranking
  • Knowledge Graph visualization of entities and relationships across the vault

AI chat & models

  • Streaming Markdown chat with conversation branching, pinning, folders, and full-text history search
  • Built-in local model (Llama 3.2 3B, GGUF via LLamaSharp with optional GPU offload) — works fully offline, out of the box
  • Ollama for any local model; opt-in cloud providers: OpenAI (GPT-4o, GPT-4o-mini, o1, o3) and Anthropic (Claude Sonnet 4, Opus, Haiku)
  • Model Manager for download/removal and per-task model selection; Hardware Advisor recommends models and GPU offload settings for your machine

Document intelligence

  • Document Comparison: structured AI comparison of 2+ documents — shared themes, key differences, unique points, Markdown export
  • Temporal Identity: belief tracking, insight harvesting, Past Self queries, and generate-as-you drafting learned from your own writing voice
  • Weekly Digest and an Analytics dashboard (usage, trends, indexing health) — computed locally from your own data

Automation & power use

  • Workflows: on-demand multi-step AI pipelines built from five step types (AI Prompt, Document Lookup, Text Transform, Conditional Branch, Output Format) in a visual builder, with full run history
  • Quick Actions, a command palette, remappable keyboard-first navigation, and a Quick Chat window
  • Plugin system (sandboxed, manifest-validated) and an authenticated local REST API

Data safety & sync

  • SQLCipher AES-256 encryption at rest; EF Core migrations; DPAPI-protected secrets
  • Backup & Restore with AES-256-GCM encrypted archives
  • Collaborative Sync between installations through an encrypted, file-based transport (any shared folder: OneDrive, Google Drive, NAS, USB) — no server, no cloud account
  • Calendar & email connectors (Outlook, Google Workspace, CalDAV / IMAP / EWS) — read-only by default, OAuth tokens stored encrypted

Platform

  • Six UI languages (en-US, de, es, fr, ja, zh-CN) — every page, dialog, and the user guide fully localized; CI enforces key parity across all locales
  • Command Console design system (DESIGN.md): a hardware-instrument UI with live status lamps and phosphor readouts, in dark (Night Shift), light (Day Shift), and high-contrast themes; WCAG-minded WinUI 3 design
  • Android mobile companion (.NET MAUI) and a browser extension for capture on the go
  • 29 navigation destinations, ~88 services, 2,775 unit tests, in-app onboarding and user guide

Documentation

Full product, architecture, and developer documentation lives under docs/:

Document Description
docs/README.md Complete product documentation — features, install, build, configuration, architecture, data storage
docs/user-guide/getting-started/quick-start.md 10-minute setup walkthrough
docs/user-guide/faq.md 100+ frequently asked questions
docs/user-guide/troubleshooting.md Solutions to common issues
docs/user-guide/glossary.md 100+ term glossary
docs/user-guide/keyboard-shortcuts.md Power-user navigation
docs/user-guide/scenarios/README.md Real-world scenarios (5 use cases)
docs/user-guide/templates/README.md Document and chat templates
docs/ARCHITECTURE.md System architecture, component diagrams, startup sequence, data-layer design
docs/DEVELOPER-GUIDE.md Services, DI, migrations, error handling, extension points
DESIGN.md The Command Console design system - tokens, typography, hardware recipes, status semantics (source of truth for all UI work)
docs/API-REFERENCE.md Public API reference
docs/PLUGIN-DEVELOPMENT-GUIDE.md Plugin development
docs/RELEASE-SIGNING.md Release provenance: keyless cosign/Rekor attestation over SHA256SUMS
CHANGELOG.md Keep-a-Changelog history across all releases

AI-crawler indexes: docs/llms.txt · docs/long-llms.txt

Release notes: v2.1.0 "Bedrock" · v2.1.0-preview.1 · v1.5.0 · v1.4.0 · v1.3.0

Build from Source

git clone https://github.com/Git-Rocky-Stack/Agent-X.git
cd Agent-X

# The solution targets x64 explicitly — a bare `dotnet build` fails with a
# win-anycpu restore error. Always pass the platform:
dotnet build -c Release -p:Platform=x64

# Run the desktop app
dotnet run --project src/AgentX.App -c Release -p:Platform=x64

# Run the test suite (2,775 tests) and the coverage gate
dotnet test tests/AgentX.Tests/AgentX.Tests.csproj -c Release -p:Platform=x64
pwsh scripts/check-coverage.ps1 -ReportOnly

Full build instructions, installer packaging (SLIM / OFFLINE Inno Setup profiles), and runtime identifiers are in docs/README.md. CI (build + tests + coverage floors, dotnet format, locale parity audit) is documented in docs/CI.md.

Contributing

Issues and pull requests are welcome. Before submitting a PR:

  1. dotnet build -c Release -p:Platform=x64 — zero warnings expected
  2. dotnet test tests/AgentX.Tests/AgentX.Tests.csproj -c Release -p:Platform=x64 — the full suite must stay green
  3. pwsh scripts/check-coverage.ps1 — coverage floors are a ratchet: they only go up
  4. dotnet format AgentX.sln — CI verifies formatting
  5. String changes must keep all six locales in key parity (CI's LocaleAudit enforces this)

License

MIT © 2026 Rocky Elsalaymeh. Built by Strategia-X.

About

Local-first AI document-intelligence desktop app for Windows (.NET 8 / WinUI 3). Turns your documents into a queryable, RAG-augmented knowledge base — HNSW vector store, hybrid semantic/lexical search, HyDE, LLM reranking. Built-in local Llama 3.2, Ollama, and opt-in OpenAI/Anthropic. No telemetry; local by default.

Topics

Resources

License

Stars

4 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors