ohdearquant

ohdearquant / khive

Public

The research knowledge graph runtime — build domain-specific KGs that grow with your work.

19
0
100% credibility
Found May 19, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

khive is a research knowledge graph runtime that gives AI assistants a structured, queryable memory for organizing research. It stores concepts, documents, notes, and the relationships between them in a simple database, allowing AI agents to read papers and automatically extract entities and connections, traverse the graph to find relationships, and search using both keywords and semantic meaning. The system uses MCP (a standard protocol for AI tool communication) to connect with AI assistants like Claude Code, requiring no complex setup beyond installing one small program and adding a configuration line. It supports multiple isolated workspaces within one database, preserves complete history through soft-delete and event logging, and offers optional task management through a Getting Things Done pack.

How It Works

1
🔬 You start researching a complex topic

You're diving deep into something like machine learning architectures, reading papers, taking notes, and trying to understand how ideas connect to each other.

2
🧠 Your AI assistant gains a memory

You connect khive to Claude Code, giving your AI assistant a structured brain that remembers every concept, paper, and insight you discuss together.

3
✨ You create knowledge that links itself

When you tell your assistant about a new concept like 'LoRA fine-tuning', it automatically creates an entry and connects it to related ideas already in your graph.

4
🔍 You search across all your research

Instead of searching through scattered files, you ask your assistant to find everything about 'parameter-efficient methods' and it returns connected concepts, not just keywords.

5
You can also manage tasks
📝
Just use the knowledge graph

Stick with organizing concepts and connections for pure research workflows

✅
Add task management

Enable the GTD pack to track what to read, what to verify, and what to write next

6
🌐 Everything stays organized and private

Your research lives in a single database file on your computer, isolated by project, with a complete history of what you created and when.

🎉 Your research assistant never forgets

You return next week to find your knowledge graph still intact, with all the connections you built ready to pick up where you left off.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is khive?

khive is a research knowledge graph runtime built in Rust that gives AI agents a typed, queryable graph for structured knowledge management. Instead of relying on vector similarity alone, it stores entities (concepts, documents, datasets, projects, people, organizations), typed edges (13 relation types), and temporal notes in a single SQLite database. The main interface is a single MCP tool called `request` that accepts a DSL for operations: you can create entities, link them via relations, search with hybrid FTS5 + vector embeddings, traverse the graph with BFS, and run GQL or SPARQL queries compiled to SQL. It ships as a single binary with no services to deploy.

Why is it gaining traction?

The hook is simplicity without sacrificing capability. Traditional knowledge graphs require Neo4j, SPARQL endpoints, and significant operational overhead. khive replaces all of that with SQLite on disk and stdio-based MCP communication. For researchers and AI developers already working with Claude Code or similar agents, this means you can express graph operations in plain DSL and the agent handles orchestration. The closed taxonomy (6 entity types, 13 relations, 5 note kinds) enforces structure that makes the graph actually useful for reasoning, not just storage. Hybrid search with reciprocal rank fusion means you get both semantic matches and exact text hits ranked by relevance.

Who should use this?

Researchers building personal or team knowledge bases who want to connect papers, concepts, and datasets without deploying a graph database. AI engineers integrating structured knowledge with agentic workflows, particularly Claude Code users who can express verbs like `search`, `traverse`, and `link` directly. Developers tired of Neo4j complexity who want a typed, queryable graph that fits in a single SQLite file. Not suitable for teams needing multi-database support (Postgres is planned but not yet available) or those requiring a visual query interface.

Verdict

khive is a clever, well-architected project with solid Rust implementation and good test coverage. The MCP-first design and SQLite simplicity are genuinely appealing. However, with only 19 stars and a 1.0% credibility score, it's early-stage and essentially a one-person effort. Treat it as an interesting experimental tool for agentic knowledge workflows rather than production infrastructure until community adoption and Postgres support mature. Watch it closely if you're building in the AI + knowledge graph space.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.