npx agentspec-cli install 4b5e1b63-38c6-4b37-a282-8e8f4cecad62 --file-type gemini-md→ GEMINI.md
# GEMINI.md — FIM One Instructional Context
## Project Overview
FIM One is an AI-powered **Connector Hub** that serves as a bridge between disjointed enterprise systems (ERP, CRM, OA, Databases) through autonomous AI agents.
### Core Architecture
- **Backend**: FastAPI (Python 3.11+) asynchronous framework.
- **Agent Core**: ReAct reasoning loops and dynamic DAG planning with concurrent execution.
- **Provider Agnostic**: Protocol-first architecture using LiteLLM to support OpenAI, Anthropic, DeepSeek, Ollama, etc.
- **RAG Pipeline**: LanceDB vector store with hybrid retrieval (FTS + RRF) and reranking.
- **Frontend**: Next.js 15+ (App Router) with shadcn/ui and Tailwind CSS 4.
- **Database**: Dual-track support for SQLite (development) and PostgreSQL (production) via SQLAlchemy and Alembic.
## Building and Running
### Prerequisites
- **Python 3.11+** with `uv` package manager.
- **Node.js 18+** with `pnpm`.
- **Docker** (optional, for sandboxed code execution).
### Key Commands
- **Full Stack**: `./start.sh portal` (Production) or `./start.sh dev` (Development with hot-reload).
- **Backend Only**: `./start.sh api` or `uv run uvicorn fim_one.web:create_app --factory --host 0.0.0.0 --port 8000`.
- **Frontend Only**: `cd frontend && pnpm dev`.
- **Database Migrations**: `uv run alembic upgrade head`.
- **Testing**: `uv run pytest`.
- **Linting/Formatting**: `ruff check src/ tests/` and `mypy src/`.
## Development Conventions
### Git & Workflow
- **Atomic Commits**: Unrelated changes must be split into separate commits.
- **i18n Automation**: Only edit English source files (`messages/en/*.json`, `README.md`, `docs/*.mdx`). Other languages are auto-translated on commit via `scripts/translate.py`.
- **Worktrees**: Agents must commit on their branch; merge via `git merge`, never file copying.
- **Migrations**: Worktree agents must NOT run `alembic upgrade head`; only write migration files. Migrations are applied after merging to main.
### Backend (Python)
- **Async First**: Use `async def` for all I/O-bound operations.
- **Type Safety**: Mandatory type hints on all public functions.
- **Migrations**: All migrations must be idempotent and compatible with both SQLite and PostgreSQL (use helpers in `fim_one.migrations.helpers`).
- **Boolean Defaults**: Use `server_default=sa.text("FALSE")` or `sa.text("TRUE")` for PG compatibility.
### Frontend (Next.js)
- **i18n**: Never hardcode English strings; use `next-intl` (`useTranslations`).
- **UI Safety**: Never `rm -rf frontend/.next` while dev server is running.
- **Admin Tables**: Follow the `admin-users.tsx` pattern: actions in a "..." DropdownMenu, Lucide icons for all menu items, destructive actions last.
- **Dialogs**: Use `AlertDialog` siblings for dirty state protection; avoid native `confirm/alert`.
- **Styling**: Prefer `focus-visible:outline-*` over `ring-*` to avoid clipping.
## Project Structure
```
src/fim_one/
├── core/ # Agent logic (ReAct, Planner, Tools, RAG)
│ ├── agent/ # ReActAgent & Hook System
│ ├── workflow/ # DAG Engine & Node types
│ ├── model/ # LLM Providers & LiteLLM integration
│ └── tools/ # Base tools & Connector adapters
├── web/ # FastAPI App, API routes, and SSE endpoints
└── db/ # SQLAlchemy models & session management
frontend/ # Next.js Application
├── src/components/ # UI components (shadcn)
├── messages/ # i18n JSON files (en/ is source)
└── public/ # Static assets
docs/ # Documentation (Mintlify format)
tests/ # Comprehensive pytest suite
```
Gemini Context: ai-platform-trainer
This file provides an overview of the `ai-platform-trainer` project, its structure, and common commands to ensure the Gemini assistant can be as effective as possible.
by AgentHub Crawler
Gemini Context: LIDS Repository
**Name:** LIDS (Live Interactive Dashboard) **Purpose:** Frontend-first sales software for Admiral Energy solar reps. **Type:** Monorepo (npm workspaces). **Stack:** React (Vite), TypeScript, Tailwind CSS, Dexie.js (Offline-first), Express (Proxy).
by AgentHub Crawler
Gemini Workspace Context: AI Agency
This document provides a comprehensive overview of the AI Agency project, its architecture, and development practices to be used for context in future Gemini conversations.
by AgentHub Crawler
Sign in to leave a comment.
Gemini CLI Integration — PMOVES.AI Ecosystem
_v1.4.0 — Unified Agentic Structure & Taxonomy_ Welcome to the PMOVES.AI production environment. As a Gemini CLI agent, you are integrated into a multi-layered, 60-agent ecosystem spanning from data persistence (Tier 1) to user interaction (Tier 7).
by AgentHub Crawler