Start here — Neural Junkie

1. Install

No Go, Node, or Rust required for the desktop app.

macOS: install with Homebrew — brew tap camronwood/tap && brew install --cask neural-junkie — or download a .dmg from Downloads.
Linux: Homebrew — brew tap camronwood/tap && brew install neural-junkie — or sudo dpkg -i the .deb from Downloads.
Windows: grab the .msi from Downloads or GitHub Releases.
Open the app. macOS: use Homebrew or see Install trust if Gatekeeper blocks.
Complete the setup wizard — pick a focus track:
- Software development — engineering specialists + IDE
- Life sciences & lab work — BiologyExpert + sequence tools (lab guide)
- Team chat & productivity — Assistant, Slack Connect, personal inbox
Choose local Ollama (recommended) or a cloud API key. macOS bundles Ollama; Windows/Linux use slim installers and the wizard Install Ollama step.

2. First chat

Open the default channel or start a DM with Assistant or a specialist.
Press ⌘K (or Ctrl+K) to open the command palette — discover slash commands without memorizing them.
Ask a real question, e.g. @Assistant what can you help me with in this workspace?
Confirm providers under Settings → AI Providers — local vs cloud usage is visible per agent.

3. Five-minute first win (developers)

Confirm Ollama or a cloud provider is connected in Settings.
Create a repo expert: /create-repo-agent /path/to/repo MyRepoExpert
Ask: @MyRepoExpert summarize the architecture and top risk areas
Start structured work: /collaborate @BackendEngineer @SecurityReviewer harden auth middleware

You should see a shared plan, specialist ownership, and an execution path with human approval before file changes.

4. Five-minute first win (lab & research)

Enable the Life sciences pack: Domain packs (⌘⇧K) → Store → Install → toggle on.
Pull models via the wizard or Model library (⇧⌘M) — OpenBioLLM 8B + qwen2.5:7b for tool routing.
DM BiologyExpert or run sequence analysis via MCP tools.
For deeper lab workflows (scan QC, TIM import, custom viewers), organizations can sideload a customer pack — see Domain packs.

What you get out of the box

Assistant — always on; command help, productivity flows, and a safety net for unanswered questions.
Routing trace — per-turn model tier and retrieval badges in chat; open the telemetry drawer to debug routing.
Project sets — group multiple repo roots; workspace scope chip shows which trees agents can see.
Runbooks v2 — persisted runbook library with run history and connector profiles.
Domain packs — install specialists for engineering, biology, CAD, AWS, incidents, music, and more.
Custom experts — any topic via /create-expert (guitar coach, compliance, internal APIs…).
Human approval — agents propose file edits and tool calls; you approve with diff preview.
Slack Connect — optional; mirror channels when your team lives in Slack.

Hardware & honesty

The app is ~15 MB; local models are not. Plan for 15–20 GB disk and 16 GB+ RAM for comfortable software-development defaults. See Hardware requirements and Known issues before you bet production work on open beta.

Next steps

Feature guides — deep dives per capability
Domain packs — catalog, install guide, and repo links
Use cases on the homepage
GETTING_STARTED.md — source builds and provider setup
GitHub Issues — bugs and feedback welcome