LLM-compiled personal knowledge base system.

Inspired by Karpathy's vision: raw documents get ingested, an LLM compiles them into a markdown wiki with summaries, backlinks, and cross-references. You can Q&A over it, generate outputs, and run health checks.

Quick Start — Plugin Mode

Add the KB plugin to any agent for interactive use:

agent = Nous.Agent.new("openai:gpt-4",
  plugins: [Nous.Plugins.KnowledgeBase],
  deps: %{
    kb_config: %{
      store: Nous.KnowledgeBase.Store.ETS,
      kb_id: "my_kb"
    }
  }
)

{:ok, result} = Nous.Agent.run(agent, "Ingest this article: ...")
{:ok, result} = Nous.Agent.run(agent, "What do we know about GenServers?")

Quick Start — Workflow Mode

For batch operations, use the workflow API:

# Batch ingest
{:ok, state} = Nous.KnowledgeBase.ingest(
  [%{title: "Article 1", content: "..."}],
  kb_config: config
)

# Health check
{:ok, state} = Nous.KnowledgeBase.health_check(kb_config: config)

Quick Start — Agent Behaviour Mode

For a KB-specialized agent:

agent = Nous.Agent.new("openai:gpt-4",
  behaviour_module: Nous.Agents.KnowledgeBaseAgent,
  plugins: [Nous.Plugins.KnowledgeBase],
  deps: %{kb_config: %{store: Nous.KnowledgeBase.Store.ETS, kb_id: "my_kb"}}
)

Architecture

The KB system has four composable layers:

1. Data model & store — Document, Entry, Link, HealthReport structs with a pluggable Store behaviour (ETS, SQLite, etc.)
2. Plugin & tools — Nous.Plugins.KnowledgeBase integrates with any agent, providing 9 tools (search, read, ingest, add_entry, link, backlinks, list, health_check, generate)
3. Workflows — Pre-built DAG pipelines for ingest, incremental update, health check, and output generation
4. Agent behaviour — Nous.Agents.KnowledgeBaseAgent for specialized KB curation and reasoning