8fe75b8c1c
- Resolve duplicate DroneStatus enum (import from @gadget/api) - Fix IAiProvider interface conflict with DB→runtime mapper - Add callId to ToolCallMessage and ChatToolCallSchema - Fix ChatTurnStats schema field name (thinkingTokenCount) - Add provider/selectedModel to ChatSession interface and model - Implement CodeSession.onSubmitPrompt() to create ChatTurn and send work orders - Add drone/chat session tracking to CodeSession - Add unit tests for CodeSession (9 tests, all passing) |
||
|---|---|---|
| .opencode/plans | ||
| docs | ||
| gadget-code | ||
| gadget-drone | ||
| packages | ||
| .gitignore | ||
| AGENTS.md | ||
| gadget.code-workspace | ||
| package.json | ||
| pnpm-lock.yaml | ||
| pnpm-workspace.yaml | ||
| README.md | ||
Gadget Code
A self-hosted Agentic Engineering Platform (AEP) — an IDE that drives autonomous AI agents to perform software engineering work on your behalf, running in your own environment.
Projects
| Package | Role |
|---|---|
gadget-code |
Web service — agentic IDE, browser UI, API server |
gadget-drone |
Worker process — runs the agentic workflow loop in workspace directories |
@gadget/ai |
Shared AI API abstraction — Ollama and OpenAI, called by both |
AI API Abstraction (@gadget/ai)
All AI API calls throughout the Gadget Code ecosystem route through @gadget/ai. No consumer code imports Ollama or OpenAI SDKs directly. No consumer code checks provider.sdk after the factory call. The shared module translates Gadget Code's internal API contract into whatever provider is configured, and translates responses back to Gadget Code's internal types.
See packages/ai/README.md for the full API reference.
Setup
pnpm install
Build
pnpm -r build # build all packages
pnpm --filter @gadget/ai build
pnpm --filter gadget-drone build
pnpm --filter gadget-code build:backend
Run
# Backend
pnpm --filter gadget-code dev
# Drone worker (in a project workspace directory)
pnpm --filter gadget-drone dev
Architecture
@gadget/ai
AI API calls are handled by @gadget/ai, which both projects depend on. This keeps all AI SDK knowledge in one place, and currently implements:
- AiApi - abstract base class for all AI APIs/SDKs
- OllamaAiApi - Ollama API implementation
- OpenAiApi - OpenAI API implementation
gadget-drone
gadget-drone is a headless process runs on end-user machines, connecting via Socket.IO to gadget-code, to receive and execute work orders for the agentic workflow loop.
At startup, gadget-drone examines process.cwd() to determine if it's a workspace directory, and if so, starts a worker process that connects to gadget-code and waits for work orders. It processes work orders in project directories in the gadget-drone workspace directory, and communicates events, status, and results to the IDE via gadget-code's web services and Socket.IO.
gadget-drone never connects directly to MongoDB or Redis — it communicates entirely through the Gadget Code API.
gadget-code
gadget-code runs on server infrastructure (MongoDB, Redis, etc.) and serves the browser-based IDE. The IDE connects to gadget-code via Socket.IO to send and receive commands in chat sessions to gadget-drone.
gadget-code can be stacked on a single host for local development, and can achieve significant scale on a single host. It can also be deployed in tiers, potentially made of clusters, and the web tier can be horizontally scaled for production use with high availability.
Libraries such as redis-adapter and redis-emitter are used for message routing and distribution in the Socket.IO library, handling the real-time message routing among gadget-code, gadget-drone, and the IDE running in the browser.