Local AI Workbench

A local-first workbench for running LLMs on your own machine with Ollama, chatting with them, and launching OpenCode against your projects with a local model already wired up. Nothing leaves your machine — no cloud calls, no telemetry.

Quickstart

1. Clone repo
2. Run setup:
   npm run setup

3. Start:
   npm run dev

4. Open:
   http://localhost:5173

npm run setup is a guided wizard — it checks your environment, tells you exactly what it's about to run, and asks before installing or downloading anything external. See docs/setup.md for what each step does, setup modes, and troubleshooting.

Internal one-command bootstrap

If you don't have the repo yet, this single command clones it, installs dependencies, and runs the setup wizard for you:

npx --yes git+ssh://[email protected]:amplimind/AImplimind.git

It prints exactly what it's about to do (repo, branch, target folder, mode, and the git/npm commands it will run), checks Ollama and optional oMLX status on your machine, asks before each step, then clones into ~/Development/AImplimind by default. The setup wizard it runs configures Ollama (all platforms) and oMLX (optional, Apple Silicon).

cd ~/Development/AImplimind
npm run dev

Windows example, with a custom target folder:

npx --yes git+ssh://git@github.com:amplimind/AImplimind.git -- --target "$env:USERPROFILE\Development\AImplimind"

Coding setup (also installs OpenCode + a coding model via the existing setup wizard):

npx --yes git+ssh://[email protected]:amplimind/AImplimind.git -- --mode coding

Check what it would do without changing anything:

npx --yes git+ssh://[email protected]:amplimind/AImplimind.git -- --check-only

Other options (pass after --): --repo <url>, --branch <name>, --target <path>, --mode <basic|coding|full> (forwarded to npm run setup), --skip-setup, --start (run npm run dev automatically once done), --yes (auto-confirm clone/install/setup — still asks before starting the dev server), --check-only (print the plan, run nothing).

Prerequisites: Node.js, npm, Git, and SSH access to the internal GitHub repo (your key added to GitHub — same as git clone [email protected]:amplimind/AImplimind.git working from your terminal already). The bootstrapper never asks for a password and never stores credentials — it just shells out to your existing git. If it can't reach the repo, it prints exactly which SSH/access checks to run yourself rather than guessing.

If the target folder already has a clone of this repo, it offers git pull instead of cloning again; if it exists but isn't this repo (or points at a different one), it stops with a clear error rather than touching it.

Honest tradeoff: because npx <git-url> (no subdirectory) only ever resolves bin from the repo root, the first run installs the full monorepo's dependencies into npm's npx cache before the bootstrapper itself can run — not just the bootstrapper's own (zero) dependencies. This is expected for the root-bin GitHub approach; subsequent runs from the same commit are normally served from that cache. We accept this for now to keep the command itself simple.

What this is

• App: NestJS API (apps/api, port 3000) + React/Vite frontend (apps/web, port 5173)
• Ollama integration: detect, install-guide, start, list/pull models
• OpenCode integration: detect, install-guide, write an Ollama-backed opencode.json, launch OpenCode in a terminal against a chosen project + model
• Chat: local chat sessions backed by a SQLite file in data/

Requirements

• Node.js 20+
• npm
• Git (for cloning/updating)
• Ollama — the setup wizard can install/guide this
• OpenCode — optional, recommended for coding workflows; the setup wizard can install/guide this too

Setup modes

npm run setup                    Guided setup (basic mode)
npm run setup -- --mode basic    Project deps + app data + Ollama + one chat model
npm run setup -- --mode coding   basic + OpenCode + a coding model
npm run setup -- --mode full     coding + extra chat/coding/embedding models
npm run setup -- --check-only    Print a status report, install nothing
npm run setup -- --yes           Auto-confirm safe project-local steps only

--yes auto-confirms folder creation and npm install. It never auto-confirms installing Ollama/OpenCode or pulling a model — those always ask first.

Windows users can also run .\setup.ps1; macOS/Linux can run ./setup.sh. Both are thin wrappers around the same node scripts/setup/setup.mjs.

Running the app

npm run dev

Starts the API (:3000) and web frontend (:5173) together. npm run start:workbench is an alias for the same thing.

Where your data lives

Everything is local, under data/ in the repo:

• data/local-ai-workbench.sqlite — chat sessions/messages
• data/local-ai-workbench/opencode/ — app-managed OpenCode config (when using "app-managed config" mode)
• logs/setup.log — setup wizard log (no secrets, no chat content, no source code)

Nothing is sent off your machine. Model inference happens through your local Ollama instance.

More documentation

• docs/setup.md — per-OS setup notes, check-only mode, troubleshooting Ollama/OpenCode