Open Source

THE LLM-FIRST WIKI

A local-first Markdown workspace with database views, hybrid search, and an interactive 3D knowledge graph. Organize it as a human, then give your agents the same context and tools.

Hybrid search 3D knowledge graph Database views Named Shards Agent skills Plain Markdown
Agent-ready by design

One vault. Shared context for humans and agents.

Tesseract skills teach coding agents how your knowledge base works — from setup and search to database structure, relations, Shards, Topics, and graph curation. Humans and agents work against the same Markdown files and .markdownvdb index.

Markdown filesShardsDatabase viewsRelationsTopicsGraph
Tesseract · Agent skills
Add Tesseract skills to this collection

Project-local context that stays with your vault.

skills ready
Claude Code .claude/skills
Detected
Codex & compatible agents .agents/skills
Recommended
Gemini CLI .gemini/skills
Available
Claude Code plugin View repository
 /plugin marketplace add geckse/tesseract-md-skills
 /plugin install tesseract@tesseract-skills

Standalone agent use requires mdvdb on your terminal PATH.

Knowledge Graph

Your ideas, connected

Explore explicit wikilinks, typed relation edges, semantic neighborhoods, and automatically discovered communities. Change layouts without losing the structure of your knowledge.

Clusters & Topics

Let Tesseract discover communities automatically, then color the graph with curated, multi-label Topics.

Every level of context

Move between collection, Shard, folder-tree, local-neighborhood, document, and chunk graph views.

Search, present & export

Find nodes instantly, animate a guided graph presentation, and export a clean visual of what you uncover.

Connected Thinking

Think in connections, not folders

Use wikilinks for ideas and typed frontmatter Relations for records. Backlinks and referenced-by views stay current, while orphan discovery shows what still needs a place in the graph.

Wikilinks

Type [[page]] to connect any note — no folders, no drag-and-drop.

Backlinks

See body links and structured reverse references, automatically updated.

Relations

Pick linked records and navigate them as graph edges or table cells.

research/alignment.md
Frontmatter 2
status active
modified 2026-04-22 14:03
Links 6
Local graph
  • Outgoing
  • Incoming
  • Bidirectional
Local graph
  • Outgoing
  • Incoming
  • Bidirectional
Alignment
Status● active Modified2026-04-22 Tags research ai-safety

Key references

  • Built on the attention mechanism from @reading/transformers
  • Related to @research/interpretability
💡 Reward modeling must preserve the preference ordering under distribution shift.

Linked notes auto-sync in the backlinks panel as you write — see @projects/tesseract.

Type '/' for commands
Related notes
  • Research
  • Technical notes
  • Wiki / journal
Related notes
  • Research
  • Technical notes
  • Wiki / journal
Notion-like editor

Write in Markdown, see it come alive

Choose Source, WYSIWYG, or Preview for every note. Work across tabs, split panes, and multiple windows with slash commands, Mermaid diagrams, highlights, tables, and embedded media — all real Markdown underneath.

Source · WYSIWYG · Preview Slash Commands Mermaid Highlights Media
Smart Search

Search that understands what you mean

Search by meaning, exact text, or both. Add frontmatter filters, keep results inside an active Shard, and expand through graph links when the answer lives one or two hops away.

Meaning-based

Finds related ideas even with different words.

Keyword

Traditional text matching when you know exactly what to find.

Filtered & scoped

Use typed properties, graph context, and Shards to narrow the result set.

Search
⌘K
5 results
  • research/alignment.md Open questions
    0.91
    Does reward modeling preserve the preference ordering under distribution shift?
  • notes/embedding-models.md Training objectives
    0.78
    Contrastive objectives as a proxy for reward — related to preference-based learning.
  • reading/transformers.md Attention, continued
    0.64
    The attention mechanism admits several reward shaping strategies post-hoc.
Related via links 2
  • projects/tesseract.md Design 1 hop
  • ideas/graph-ui.md Notes panel 2 hops
Graph context
  • Your query
  • Direct result
  • 2-hop via link
Graph context
  • Your query
  • Direct result
  • 2-hop via link
Structured knowledge

A wiki that behaves like a database

Turn folders into live database views without moving your data out of Markdown. Add structure when it helps, keep plain files underneath, and let agents query the same records.

Clients · Table Acme Corp.md
FilterSortGroup
All clientsActiveBy region+
ƒx Formula LookupΣ Rollup3 records · €284,350 total
Database Table Views

Folders become databases

Filter, sort, group, save views, and edit cells inline while every row remains a Markdown document.

Properties

Typed frontmatter

Use dates, selects, JSON, files, and relation pickers with schema-aware editing in documents and tables.

Computed

Formula, Lookup & Rollup

Create read-only computed fields that materialize back into Markdown and stay available to every tool.

Relations

Real links between records

Connect folders like tables, follow reverse references, and see relation edges directly in the graph.

Named Shards

Focus a large collection without splitting it apart.

A Shard is a named recursive folder lens with focused navigation, search, tables, graph analysis, and independent Topics — all backed by the collection’s shared index.

Knowledge2,418
└─Research684
   └─AI Safety129
Active scope Knowledge › Research
Scoped searchLocal TopicsShard graph
A complete workspace

More than a Markdown editor

Tesseract brings the surrounding tools into the same local workspace, so you can write, organize, automate, and inspect without stitching together a cloud stack.

Obsidian-compatible

Open existing vaults and let Tesseract import and safely sync Topics from tags and graph color groups.

Terminal + first-class CLI

Run commands beside your notes or automate the same collection through mdvdb’s structured JSON interface.

Live indexing activity

See watcher events, ingest progress, diagnostics, and computed-field updates without losing your editing flow.

Light, dark, and yours

Choose light or dark mode, a custom accent color, and a harmonic graph palette that follows your workspace.

Files and media

Attach files as typed properties, paste images directly, preview PDFs and media, and keep assets collection-local.

Bring your embedding provider

Use OpenAI, OpenRouter, Gemini, Azure OpenAI, AWS Bedrock, Hugging Face, Ollama, or a custom endpoint.

Your Data

Your files and index stay under your control.

Tesseract keeps your Markdown, assets, schema, and index on your machine. Use Ollama for fully local embeddings, or connect only the cloud provider you explicitly choose.

Plain Markdown

Standard .md files. No vendor lock-in. Take your notes anywhere.

Local Index

Search indexes and configuration stay beside your collection. No account required.

Your Embeddings

Run locally with Ollama or bring credentials for the provider you select.

Open Source

MIT licensed. Audit the code. Contribute on GitHub.