VanillaGraph -- v0.1.0
VanillaGraph · Open source · MIT · Local-first

A local knowledge graph
where AI agents do the filing.

VanillaDB builds tools for agent memory and databases. VanillaGraph is our first. Drop in documents, agents extract concepts and propose wiki articles, you approve. Your knowledge graph compounds entirely on your machine.

  • Four-agent pipeline: ingest → analysis → proposal → fileback
  • Knowledge graph with typed wikilinks and hub detection
  • MCP server: use your vault as a tool in Claude Desktop
  • OpenAI, Anthropic, OpenRouter, or fully-local Ollama
  • Hybrid BM25 + semantic search across your wiki
  • Tauri desktop app for Windows, macOS, Linux
View on GitHub Read the docs
Open source · MIT licensed · Intentionally vanilla

Database infrastructure
for the agentic era.

VanillaDB builds open-source database components designed for agents: the infrastructure layer for how agents read, write, and remember. Our releases are intentionally minimal, built to be forked, adapted, and made your own.

View on GitHub Explore VanillaGraph →
Pipeline

Four agents,
one pipeline.

Documents go in. Structured wiki articles come out. Each agent has a narrow, auditable job. Nothing is written to your vault until you approve it. The whole pipeline is open-source. Customize every step to fit your workflow.

vanilla-pipeline
$ vanilladb ingest attention-is-all-you-need.pdf
 
[01 · Reader]Parsing PDF · 847 lines extracted
[01 · Reader]Topics: attention, transformers, nlp (+9 more)
 
[02 · Context]Scanning vault · 142 existing notes
[02 · Context]Gaps: 3 · Potential links: 8
 
[03 · Writer]Drafting wiki articles...
[03 · Writer]wiki/Transformer Architecture.md✓
[03 · Writer]wiki/Self-Attention Mechanism.md✓
[03 · Writer]wiki/Positional Encoding.md✓
 
[04 · Commit]Review approved · writing to vault
[04 · Commit]graph +node Transformer Architecture
[04 · Commit]graph +node Self-Attention Mechanism
[04 · Commit]graph +node Positional Encoding
[04 · Commit]+3 wiki files · +3 graph nodes · +11 edges ✓
knowledge-graph
Neural Nets Lang. Models Embed- dings Atten- tion Transformer Arch. Self- Attention Positional Encoding
+3 nodes · +6 edges
Roadmap

What we're
building next.

VanillaGraph is live at v0.1. Here's what's coming next. All open-source, all intentionally vanilla. Every tool we ship is designed to be understood, extended, and made your own.

Memory Layer
A structured backend for agent memory that persists context across sessions. Standard interfaces, fully local.
In research
Data Normalization
Standard ingest pipelines for PDFs, transcripts, and mixed-format records into formats agents can reliably query.
In research
Memory Benchmarking Suite
Tooling to properly evaluate agent memory systems: recall accuracy, context freshness, and retrieval performance under realistic workloads. Open-source and extensible.
Planned
Simplicity-first Primitives
Database and architecture components that solve complex agent system problems through minimal, composable abstractions. One job, simple interface, fully auditable.
In research
Community

A place for builders
using local AI tools.

VanillaDB is MIT-licensed and developed in the open. Join the Discord to be part of the conversation, from first questions to deep contributions.

MIT
License, use it anywhere
v0.1
VanillaGraph, live now
0→∞
Local-first, no accounts

We discuss roadmap decisions publicly, welcome contributions, and share early builds with community members first. If you're building with VanillaDB, we want to hear about it.

VanillaDB Discord
  • Get answers to setup and usage questions
  • Inquire about contributing or collaboration
  • Get early access to new tools before release
  • Chat with other local-AI builders
Join Discord
github.com/sanalvre/VanillaGraph
Source, issues, pull requests