Vivid Agent is an extensible, open‑source framework for building autonomous AI agents in TypeScript. It combines persistent memory, tool execution, and a flexible AI engine to create agents that can run tasks with minimal supervision—or respond to user requests on demand.

🚧 Actively developed – we're looking for contributors to help shape the future of autonomous agents!

• Memory system – Long‑term (Markdown file) and short‑term (in‑memory context) memories.
• Tool framework – Easily add tools (file manipulation, command execution, etc.) with sandboxing.
• AI engine abstraction – Use NVIDIA's models, a mock engine for testing, or plug in your own.
• Workspace isolation – All files and logs are contained in a dedicated directory.
• Logging – Coloured console output + file logs with multiple levels.

We aim to build a reliable, self‑hosted agent that can:

• Run autonomously in the background, performing scheduled tasks.
• Learn from interactions via long‑term memory.
• Be extended with custom tools and skills by the community.

git clone https://github.com/vividorg/agent.git
cd agent
npm install
npm run build
chmod +x dist/index.js
npm link

Start the service and open an interactive session:

vivid service         # start the agent service
vivid tui             # open interactive prompt (/exit to quit)

Or send a one-shot prompt:

vivid tui -m "Hello, what can you do?"

Without npm link: use npm run service and npm run tui instead of vivid service / vivid tui.

Default service URL is http://127.0.0.1:3100, configurable via --url or VIVID_SERVICE_URL.

1. Start llama.cpp with an OpenAI-compatible endpoint:

./llama-server -m /path/to/model.gguf --host 0.0.0.0 --port 8080

2. Copy .env.example to .env and configure:

AI_ENGINE=llama
LLAMA_BASE_URL=http://127.0.0.1:8080
LLAMA_MODEL=local
LLAMA_MAX_TOKENS=4096
NVIDIA_API_KEY=nvapi-key   # not required for llama engine

3. Run:

vivid service
vivid tui -m "Hi there, what can you do?"

docker compose up -d --build
vivid tui -m "Hi there, what can you do?"

Agent data is persisted in ./data/ (VIVID_HOME=/data inside the container).

vivid service &       # simple background start

Or with PM2 for process management:

npm run pm2:start
pm2 status
npm run pm2:stop

Storage path is configurable via VIVID_HOME (default: ./.vivid in the working directory).

Development happens in the open, and we'd love your help!

• Discord – real‑time discussions, support, and ideas.
• GitHub Issues – report bugs or suggest features.
• Contributing guide – look for good first issues to get started.