Skip to content

Spikuit

Knowledge, curated through conversation.

No preprocessing. No chunking pipelines. No metadata schemas to maintain. Just add your documents and talk with your agent.

Spikuit (spike + circuit, pronounced /spaɪ.kɪt/) is a personal knowledge system where an AI agent handles the hardest parts of knowledge management — ingestion, structuring, and maintenance — through dialogue.

Traditional RAG systems break down at data curation: chunking, tagging, connecting, keeping things fresh. Spikuit solves this with Conversational Curation — the agent curates your knowledge base as you talk to it.

Three skills, one loop

/spkt-ingest — Talk it in.

Feed articles, notes, or URLs into your Brain. The agent chunks content, discovers relations, and builds your knowledge graph — you just talk.

You: /spkt-ingest
     Summarize this for my brain: https://arxiv.org/abs/1706.03762

Agent: Added 8 neurons from "Attention Is All You Need".
       6 synapses created, source linked for citation.

/spkt-qabot — Ask it back.

Query your Brain with natural language. Answers include source citations. Retrieval quality improves with every conversation — unhelpful results are automatically penalized, helpful ones are boosted.

You: /spkt-qabot
     How does multi-head attention differ from single-head?

Agent: Multi-head attention runs multiple attention functions in parallel...
       Sources:
       - [Attention Is All You Need](https://arxiv.org/abs/1706.03762)

/spkt-tutor — Let it teach you.

An AI tutor built on your knowledge graph. It detects prerequisites, adapts difficulty, and gives feedback on mistakes — not just "correct" or "wrong".

You: /spkt-tutor

Tutor: Let's start with Functor — it's a prerequisite for the other two.
       [teaches → quizzes → gives feedback → re-explains weak areas]

How It Works

  1. Smart scheduling — each concept has its own review timing (FSRS)
  2. Activation spreading — reviewing one concept nudges related concepts closer to review. Connections used together get stronger.
  3. Search optimization — results ranked by relevance × memory strength × graph centrality. Feedback improves ranking over time.

Quick Start

# Install
pip install spikuit

# Initialize a brain (interactive wizard)
spkt init

Then, from your Agent CLI (Claude Code, Cursor, Codex):

/spkt-ingest    → Talk it in. Curate knowledge through conversation.
/spkt-qabot    → Ask it back. Get cited answers from your knowledge graph.
/spkt-tutor    → Let it teach you. Study with an AI that adapts to your level.

Or use spkt commands directly:

spkt source ingest ./papers/ -d cs --json     # Ingest a directory with metadata
spkt retrieve "query" --filter domain=math
spkt diagnose                                # Brain health check
spkt consolidate                             # Sleep-inspired graph optimization
spkt export -o brain.json --format json
spkt visualize

Documentation

License

Apache-2.0