VAI is a lightweight AI agent runtime and tooling suite built in V.

(中文版本: README.zh-CN.md)

• Usage (English): docs/USAGE.md
• 使用指南（中文）: docs/USAGE.zh-CN.md

• High performance, single binary build
• Multi-provider LLM support (OpenAI, OpenRouter, Ollama, …)
• Event-driven message bus and Agent loop (context → LLM → tools → memory)
• Multi-agent collaboration (hub, routing, task scheduling)
• Skills/tool execution framework (builtin, dynamic, Markdown with YAML frontmatter)
• Layered memory (conversation store, daily notes, long-term storage)
• Cron and heartbeat services for scheduled and proactive tasks
• Web dashboard + REST API
• CLI console and interactive chat mode

# in repo root
v -prod .
./vai --help

Initialize configuration and workspace (creates ~/.vai/config.json and ~/.vai/workspace/):

./vai onboard

Then edit ~/.vai/config.json to add API keys and model settings.

./vai web

Then open http://localhost:8080.

Pick one provider (or set in ~/.vai/config.json after vai onboard):

# OpenAI
export OPENAI_API_KEY="..."

# or OpenRouter
export OPENROUTER_API_KEY="..."
export VAI_DEFAULT_MODEL="gpt-4o-mini"  # optional

# or local Ollama
ollama serve

vai/
├── agent/       # Agent loop, context builder, hub/coordination
├── bus/         # Event-driven message bus (events, queue)
├── config/      # Config schema and loader (~/.vai/config.json)
├── cron/        # Cron job types and scheduler
├── cli/         # CLI console and commands
├── gateway/     # Platform adapters (Telegram/Discord/WhatsApp/DeBox/…)
├── heartbeat/   # Heartbeat service for proactive tasks
├── llm/         # LLM clients/providers
├── memory/      # Store, persistent, daily_notes, long_term, embeddings
├── planner/     # Planning strategies
├── protocol/    # Message protocol
├── sandbox/     # Sandbox execution (simplified)
├── scheduler/   # Task scheduler
├── skills/      # Registry, builtin, dynamic, markdown skills
├── workspace/   # Workspace manager and templates (~/.vai/workspace/)
└── web/         # Web UI server + REST API + static dashboard

See arch.md for architecture details and data flow.

The dashboard page (web/static/index.html) calls /api/agents and /api/status. At the moment, the frontend expects a {success,data} wrapper, while many API endpoints return plain JSON arrays/objects. See the usage guide for details and workarounds.