Getting Started

The PaglaAI launcher starts local AI from one place. Install it, add an API key or two, and dispatch work to Gemini, Kilo Code, Claude Code, or Qwen Code without managing them by hand. The launcher is a standalone project — its source lives in the sibling PaglaAI core repository, not inside this design site.

Installation

Run the setup script to create a virtual environment and install the Python dependencies.

Windows · PowerShell
$ setup.bat
# creates .venv and installs google-genai + python-dotenv

Then fill in .env from .env.template. Keys are comma-separated lists for multi-account failover:

.env
GEMINI_API_KEY=k1,k2
ANTHROPIC_API_KEY=sk-ant-xxx   # optional

Quick start

Launch the core. On first run the onboarding wizard checks your keys and classifies them live.

Mode prefixes select the agent: code: → Kilo Code, build: → Claude Code, auto: → Qwen Code, no prefix → native Gemini chat.

Concepts

PaglaAI treats every agent as a dispatcher target. One launcher, several backends, a single prompt line. The shadow reflector wraps every CLI run — recording to shadow/, capping output, and flagging empty claims.

Tutorials

See the API reference for a worked chat completion example, then explore the architecture section for how the launcher wires dispatch, failover, and the shadow reflector together.

Architecture

The launcher wires three layers together: dispatch (which agent handles the request), failover (key rotation across accounts), and the shadow reflector (which records every run and flags empty claims).

FAQ

  • Do I need a cloud API key? No — local models run fully offline; keys only enable cloud routing.
  • Where do run logs live? The shadow reflector writes to shadow/ inside the project directory.
  • How do I add another provider? Add its keys to .env; the dispatcher detects them and routes automatically.