Neuropipe

Neuropipe is a Windows-first desktop workspace for building local AI automations. It combines a Blueprint-style visual editor, local and compatible LLM providers, scheduled and button-triggered pipelines, chat, reports, and operational metrics in one Wails application.

Everything is designed to stay on your machine by default. Pipelines, reports, execution history, metrics, model metadata, and settings are stored in the local Neuropipe workspace. Secrets are kept separately in the Windows vault and are redacted from previews and logs.

Install on Windows

The normal way to install Neuropipe is to download the installer from the latest Neuropipe release. It installs the desktop app without requiring Go, Bun, or Wails.

You can also browse every release to download a specific version or the standalone executable.

What you can build

• Button boards, cron schedules, file watchers, hotkeys, chat, and HMAC-protected local webhooks.
• Blueprint-style pipelines where exec pins control actions and typed data pins resolve values on demand.
• LLM workflows using Ollama, managed llama.cpp, Anthropic, or any OpenAI-compatible endpoint — with several providers configured side by side.
• HTTP, local files, PowerShell, processes, clipboard, notifications, Git, reports, reusable functions, and AI agents.
• Markdown reports, local chat conversations, and a privacy-safe metrics dashboard.

Getting started

1. Open Settings → Provider and configure one or more providers (pick a default; add models per provider). For managed llama.cpp, select a runtime and an installed GGUF model.
2. Create a pipeline, add a trigger, then connect exec and data pins in the editor.
3. Configure its nodes, save the draft, and publish the revision after reviewing requested capabilities.
4. Run it from the editor or the Trigger board. Schedules and webhooks need an explicit trust grant before they run unattended.

The built-in Documentation tab contains detailed guides and a reference page for every available node.

Build from source

To develop or build Neuropipe yourself, install Go 1.26+, Bun, and the Wails v3 prerequisites. Then install the Wails v3 CLI and run:

git clone https://github.com/FlameInTheDark/neuropipe.git
cd neuropipe
go install github.com/wailsapp/wails/v3/cmd/[email protected]
wails3 dev

To create a production Windows build:

wails3 generate syso -arch amd64 -icon build/icon.ico -manifest build/wails.exe.manifest -info build/info.json
go build -tags production -trimpath -ldflags "-s -w -H windowsgui" -o build/bin/neuropipe.exe

The executable and installer are written to build/bin.

Development checks

cd frontend
npm install
npm run check
npm run build

cd ..
go vet ./...
go test -race ./...

The remote-executor gRPC contract is generated with protoc; regenerate it after editing internal/proto/executor/v1/executor.proto:

task gen:proto

DuckDB database support is opt-in on Windows (its static libs need an MSVC-compatible link step). Build with -tags duckdb to include it:

go build -tags duckdb -o build/bin/neuropipe.exe

The standalone pipeline executor (Windows/Linux/macOS, no cgo) is built with:

task executor:build

Its CLI uses serve to start the daemon, auto-generates and saves a shared token on first start (printed exactly once for copying into Neuropipe), and ships a few tools:

neuropipe-executor serve --token <value>   # explicit token; also --config/--data-dir/--listen
neuropipe-executor token generate          # rotate the shared token, printed once
neuropipe-executor status                  # effective config + local pipeline/run counts
neuropipe-executor --version

Releases

Pull requests and pushes to main run the frontend type/build checks plus Go vet and race tests. Conventional commits on main create a semantic GitHub release. A second workflow builds the tagged Windows executable and NSIS installer, then attaches both to that release.

Use feat: for a minor release, fix: for a patch release, and add a breaking-change footer for a major release.