conexus 1.10.2

Self-hosted semantic search and knowledge management for LLM-driven development

Nexus

Nexus is a support framework and tooling foundation for systematic development with evolving semantic knowledge. It gives teams and individual developers a way to index what they build, search what they know, and maintain coherent direction as projects grow in complexity — especially under the speed and pressure of LLM-driven agentic coding.

Three layers, each useful on its own, each building on the last:

1. Semantic search and repository indexing (nx) — Index any repo: code is chunked with tree-sitter AST parsing, prose with semantic markdown splitting, PDFs with layout-aware extraction. Search across all of it with a single command. Three storage tiers let you start local with zero API keys and add cloud search when you're ready.

2. A Claude Code plugin (nx/) — 14 agents, 27 skills, session hooks, slash commands, and two bundled MCP servers. Agents access all three storage tiers via structured MCP tools (no Bash dependency), search indexed code before proposing changes, and coordinate through standard pipelines (plan, implement, review, test). Works with the CLI; does not require RDR.

3. A structured decision framework (RDR) — Research-Design-Review documents: the traction control for agentic development. Each decision is a short document with classified evidence (Verified, Documented, or Assumed). Write one, build it, learn something, write another. Nexus keeps the growing corpus searchable and navigable. Fully optional.

Use just the CLI. Add the plugin. Adopt RDR later, or never. Each layer amplifies the ones below it but none requires the ones above.

Quick Start

uv tool install conexus          # install the nx CLI from PyPI

nx config init                   # configure API keys
nx doctor                        # verify setup

nx index repo .                  # index current repo
nx search "authentication flow"  # semantic search
nx search "auth" --hybrid        # semantic + git frecency

Scratch and memory commands work with zero API keys. Cloud search requires ChromaDB and Voyage AI accounts — both offer free tiers that cover all typical Nexus usage. See Getting Started.

Repository Indexing

nx index repo is the core of Nexus. Point it at any git repo and it:

1. Walks tracked files via git ls-files (respects .gitignore)
2. Classifies each file by extension — code, prose, or PDF
3. Chunks code with tree-sitter AST parsing (31 languages, 52 file types) and prose with semantic markdown splitting
4. Embeds each chunk with a purpose-built Voyage AI model
5. Routes results to separate collections: code__<repo>, docs__<repo>, and rdr__<repo> for RDR documents
6. Computes git frecency scores so recently-touched files rank higher in hybrid search

Auto-discovers RDR documents in docs/rdr/ and indexes them into a dedicated collection. Stable across git worktrees. Configurable per-repo via .nexus.yml. See Repo Indexing.

The CLI

Three storage tiers with increasing durability:

Tier	Storage	Network	Use
T1 — scratch	In-memory ChromaDB	None	Session-scoped working notes
T2 — memory	Local SQLite + FTS5	None	Per-project notes that survive restarts
T3 — knowledge	ChromaDB cloud + Voyage AI	Required	Permanent semantic search

Every command targets one or more tiers:

Command	Tier	Description
nx search	T3	Semantic and hybrid search
nx index	T3	Index code repos, PDFs, and markdown
nx store	T3	Store, export, and import knowledge
nx memory	T2	Per-project persistent notes
nx scratch	T1	Ephemeral session scratch pad
nx collection	T3	Inspect and manage cloud collections
nx config	—	Credentials and settings

RDR: Research-Design-Review

Agentic coding is like driving on slick ice — fast, powerful, and easy to lose control of. RDR is the traction control: a short document where you state the problem, record what you know, describe the plan, and note what you rejected. It gives humans a structured way to steer LLM-driven development and feed discoveries back into the next decision.

Each finding is classified so readers know what is solid and what is a guess:

Classification	Meaning
Verified	Confirmed via source code search or working spike
Documented	Supported by external documentation only
Assumed	Unverified — flag it if your design depends on it

RDRs are iterative, not waterfall. You write one, build it, learn something, and write another. A real project might produce 20+ RDRs over its lifetime — foundation decisions, mid-project pivots when assumptions break, performance fixes from real usage, quality refinements from actual data. Each one builds on what you learned implementing the last.

That many design documents create an information management problem. Nexus handles it: every RDR is semantically searchable the moment it is committed, metadata is queryable without parsing markdown, and agents receive prior-art context automatically. The corpus stays navigable as it grows. See RDR Overview.

The Plugin

The nx/ directory is a Claude Code plugin. Install via the marketplace:

/plugin marketplace add Hellblazer/nexus
/plugin install nx@nexus-plugins

The plugin provides:

• 14 agents — code review, debugging, architecture planning, research synthesis, strategic planning, and more
• 27 skills — RDR workflow, TDD discipline, brainstorming gates, nexus CLI reference
• Session hooks — auto-initialize scratch, surface T2 memory context, health-check dependencies, prime beads
• Slash commands — /research, /create-plan, /review-code, /rdr-create, /rdr-accept, etc.
• Standard pipelines — feature, bug, and research workflows with built-in review gates
• Two bundled MCP servers — nexus (8 storage tier tools for agents) and sequential-thinking, both via .mcp.json

Each agent runs on a model matched to its task: opus for complex reasoning, sonnet for implementation, haiku for utility. The plugin integrates with Beads for task-level tracking — session hooks prime bead context, RDR close decomposes decisions into beads, and branch naming ties back to bead IDs. See nx/README.md for the full plugin documentation.

Documentation

Document	What it covers
Getting Started	Install, configure, first index and search
CLI Reference	Every command, every flag
Storage Tiers	T1/T2/T3 architecture and data flow
Memory and Tasks	T2 memory, beads integration, session context
Repo Indexing	Smart file classification, chunking, frecency
Configuration	Config hierarchy, .nexus.yml, settings
Architecture	Module map, design decisions
Contributing	Dev setup, testing, code style

RDR (Research-Design-Review): Overview · Workflow · Nexus Integration · Templates · Project RDR Index

Prerequisites

• Python 3.12+, uv (for install), git (for repo indexing)
• ChromaDB cloud + Voyage AI for T3
• ripgrep for hybrid search

License

AGPL-3.0-or-later. See LICENSE.