Installation

What to install

Three shapes. Pick one based on how you'll use TAI.

As a CLI

The most common shape. Installs the tai command globally.

bash

npm install -g @tailored-ai/cli

Run tai init once to create ~/.tailored-ai/config.yaml, then use tai from any working directory. The -c flag selects a different config and uses its parent directory as that instance's home. Installation works with any package manager:

bash

pnpm add -g @tailored-ai/cli
yarn global add @tailored-ai/cli
bun add -g @tailored-ai/cli

As a project dependency

If you want the agent's version pinned to your project's lockfile:

bash

npm install --save-dev @tailored-ai/cli

Run via your package manager's script runner: npx tai, pnpm tai, yarn tai, bunx tai. TAI still reads ~/.tailored-ai/config.yaml unless you pass -c ./config.yaml or set TAI_HOME.

As a library you embed

When you want the runtime inside your own Node service rather than running the tai command, depend on the libraries directly:

bash

npm install @tailored-ai/core @tailored-ai/server better-sqlite3

See @tailored-ai/core for the embed API and @tailored-ai/server for the HTTP surface.

System requirements

Node.js 20 or newer. Node 18 reached EOL in April 2025. TAI declares engines.node: ">=20".
A C toolchain only if better-sqlite3 doesn't have a prebuilt binary for your platform. Prebuilt binaries cover Linux, macOS, and Windows on x64 and arm64. If install fails: install python3 and a compiler (Xcode CLT on macOS, build-essential on Debian/Ubuntu, Visual Studio Build Tools on Windows).

Pick a provider

TAI doesn't ship a model. Pick the shape that matches where your model runs.

Local or self-hosted (Ollama, vLLM, LM Studio)

The built-in openai_compatible provider talks to any server that speaks the OpenAI chat API. For Ollama on a developer laptop:

bash

brew install ollama        # or .deb / .rpm / install script
ollama serve &
ollama pull llama3.2

yaml

providers:
  openai_compatible:
    baseUrl: http://localhost:11434/v1
    defaultModel: llama3.2

agent:
  defaultProvider: openai_compatible

For vLLM swap the baseUrl for http://127.0.0.1:8000/v1. An apiKey is optional and only sent when set.

Hosted vendors

Hosted providers ship as plugins. Install the one you want, configure its providers block, and select it:

bash

tai plugin install @tailored-ai/provider-anthropic

yaml

providers:
  anthropic:
    apiKey: ${ANTHROPIC_API_KEY}
    defaultModel: claude-haiku-4-5

agent:
  defaultProvider: anthropic

.env alongside config.yaml:

ANTHROPIC_API_KEY=sk-ant-…

The exact credentials vary by provider. OpenAI, OpenRouter, and DeepSeek use API keys; Bedrock uses the standard AWS credential chain. Configure a model provider compares the available adapters and links to each provider's field-level reference.

Multi-provider

You can also declare several providers and let each agent pick its own. See Configuration → Multi-provider.

Optional dependencies

A few tools have peer dependencies installed only if you turn them on. Each tool fails with a clear "run npm install …" message at first use if the peer is missing.

Tool you turn on	Peer to install
`browser`	`npm install playwright && npx playwright install chromium`
`browser_mediator`	`npm install @tailored-ai/browser-mediator playwright && npx playwright install chromium`
`md_to_pdf`	`npm install md-to-pdf`
PDF parsing (used by document tools)	`npm install pdf-parse`
OCR on scanned PDFs and images	`npm install tesseract.js`
Trusted-actions Amazon purchases	`npm install @tailored-ai/trusted-actions`; runs in its own Docker container

Where things live on disk

By default TAI keeps one instance under ~/.tailored-ai/, independent of the directory where you run the command.

~/.tailored-ai/
├── config.yaml          # required
├── .env                 # optional, for ${VAR} interpolation
├── agent.db             # SQLite; created on first run
├── data/                # embeddings, knowledge base; created on first run
├── plugins/             # packages installed with tai plugin install
└── workflows/           # optional; YAML workflow files

To keep another isolated instance, set TAI_HOME=/path/to/dir or pass tai -c /path/to/dir/config.yaml. TAI keeps that instance's config, database, plugins, keys, and generated data under the selected home.

Verifying the install

bash

tai --help            # CLI help
tai --list-agents     # agents your config defines
tai -m "Say hi"       # one-shot message; tests the provider connection

If --list-agents works but tai -m "…" hangs or errors, the LLM provider isn't reachable. Check that Ollama or vLLM is running, or that your API key is set.

Uninstall

bash

npm uninstall -g @tailored-ai/cli

~/.tailored-ai/ still contains your config, database, plugins, and generated data. Remove that directory separately if you do not want to keep the state.