IDE v4 — build the IDE you actually own

If Elon Musk bought your favorite IDE, come join our open-source effort to build a better one.

I'm shipping Neural Junkie IDE v4 — not as a reaction tweet, but as years of local-first, multi-agent workspace work reaching the depth developers expect from a serious editor.

Why this moment matters

AI coding tools are consolidating fast. When a beloved IDE becomes an acquisition target, developers rightly ask:

Who controls the roadmap?
Where does my code go by default?
Can I still run models locally?
What happens to the workflow I built my team around?

Open source doesn't solve everything — but it answers the ownership question clearly.

What v4 adds

Full Monaco LSP (local workspaces)

We moved from LSP-lite (one-shot gopls check) to persistent language servers behind a hub WebSocket:

Hover in the editor (live in Monaco)
Foundation for completion, go-to-definition, rename
Go, Rust, Python via gopls, rust-analyzer, pyright-langserver

Your hub spawns and owns the language server process. The desktop speaks JSON-RPC over WebSocket — same pattern as VS Code, without sending your repo to our cloud (there is no our cloud).

Remote SSH via `nj-remote`

Workspaces are no longer "a path on the laptop only."

We added WorkspaceBackend — a pluggable FS/exec layer — and a small nj-remote sidecar you run on EC2, a dev box, or anywhere SSH reaches:

nj-remote -root ~/myapp -addr :19876 -token "$SECRET"

The desktop hub routes file operations, git, LSP, and implementation-session verify steps through the backend. Agents stay local; your code stays where you put it.

Dev containers

Repos with .devcontainer/devcontainer.json get an attach plan from the hub. Run nj-remote inside the container after devcontainer up — same sidecar protocol, container as the host.

tree-sitter symbols

Symbol search upgrades from regex to tree-sitter when the CLI is on PATH, with regex fallback. ⌘⇧O keeps working; nested methods stop hiding.

What didn't change (on purpose)

Human approval on file changes — proposals and diffs, not silent writes
Multi-agent specialists — not one overworked chatbot
BYOM — Ollama, Claude, GPT, LM Studio, Hugging Face — your keys, your machine
Local-first hub — loopback by default

Join the build

Neural Junkie is open source (see the repo license). v4 is the invitation:

Download the beta — macOS, Windows, Linux
Try IDE layout + Software development pack
Open an issue if remote/LSP breaks on your stack
Send a PR — WorkspaceBackend, LSP, sidecar, scenarios

If the IDE you loved now has a new owner, you don't have to wait for their roadmap.

Come build the next one with us.

https://github.com/camronwood/neural-junkie

Download CTA: https://camronwood.github.io/neural-junkie/download.html

Suggested post date: After v1.2.0-beta.2 tag (CI + remote LSP dogfood).