Stop copy-pasting rules between coding agents: one Git repo for project knowledge

If you use coding agents across several projects, you probably maintain the same instructions in several places.

A rule gets copied into CLAUDE.md in one repo and AGENTS.md in another. The copies drift. One gets fixed, the others don't. Important architecture decisions stay buried in documentation the agent never sees. And you keep re-explaining the same project background at the start of every session.

I built Nexora to make that easier.

One knowledge repo, shared across agents

Keep your rules, skills, and decisions in one Git repository, versioned and reviewed like code. Then reuse them with Claude Code and other coding agents.

Nexora doesn't dump everything into the prompt. It selects the knowledge that applies to the project and the files being changed, and every item comes with:

• its source
• its version
• the reason it was selected

You can see which rules were selected, where they came from, and why they apply.

Required rules don't depend on search

You can add documentation files or Confluence pages and use vector search for RAG. Or keep your existing documentation service: the agent calls it directly, and Nexora supplies guidance on how to use it.

But here's the part I care about most: required rules are selected separately from search results. Their inclusion is deterministic, based on project, file, and agent, so a mandatory rule never gets dropped because it scored low on relevance.

If the documentation search is down, you still get your project rules, plus a clear note that the docs were unavailable.

Native instructions and skills

Nexora can generate native instructions and skills for Claude Code and Codex. With native mode enabled (resolve --native, or native: true over MCP), it checks that those files are current and returns references instead of repeating their content.

Nexora keeps context focused: it matches knowledge by project, agent, and file path, then retrieves a limited number of documentation passages for the task. With native mode enabled, rules and skills already in verified, current agent files are referenced instead of repeated. Applicable mandatory rules are always kept.

Works offline

Local knowledge and keyword search need no database and no API key. It's a single Go binary.

Try it locally:

# macOS / Linux
curl -fsSL https://raw.githubusercontent.com/acierto/nexora/main/scripts/install.sh | sh
nexora init --project my-api
$EDITOR knowledge/rules/project.md
nexora validate
nexora sync --agent claude # or: --agent codex
nexora resolve --agent claude --file src/auth/service.go

Use the generated native files with your agent, or connect nexora mcp for on-demand context lookup.

Sharing across projects: point each project's configuration to a local checkout of your shared knowledge repo. Pull changes and sync again when you want to update the generated agent files.

Status

Nexora is in early development. Knowledge selection, docs and Confluence indexing, pgvector, the MCP server, and native sync for Claude Code and Codex work today. Shareable knowledge packs and more connectors are on the roadmap.

Who it's for

Nexora is useful even for one repo if you want your agent to use project rules alongside documentation retrieved through RAG. Across several repos, it also gives you one place to maintain shared rules, skills, and decisions.

I'd love feedback, especially on how you manage agent instructions across projects today. Drop a comment 👇

👉 Try Nexora on GitHub
📚 Read the docs