Tailored AI

Tailored AI

Tailored AI is a runtime for LLM agents that do work on a schedule. It runs on a laptop, on a Raspberry Pi, or on a server. The agents have access to your shell, your files, your inbox, and your calendar. They reach you in the terminal, in a browser, on Discord, or through your own service.

What a minute with TAI looks like

yaml
# config.yaml
provider:
  type: openai
  model: gpt-4o-mini
  apiKey: ${OPENAI_API_KEY}

agents:
  reporter:
    instructions: |
      You write standups. Three bullets: yesterday, today, blockers.
      Be specific. Don't pad.
    tools: [exec, write]

tools:
  exec:  { enabled: true, allowedCommands: [git, date] }
  write: { enabled: true }
bash
tai -a reporter -m "Run git log for the past day and write today's standup to standup.md"

Twenty seconds later, standup.md exists and reads like a person wrote it. Switch provider.type to ollama and the same config runs offline. Add a cron block and it runs every morning at eight without you. Add a Discord section and it DMs you the result.

What this is good for

  • Cron jobs that think. Read your calendar at 7am and prep a briefing. Watch your inbox for invoices. Generate a release-note draft when a semver tag is pushed.
  • Multi-agent workflows. A coder agent writes the diff in a fresh git worktree, hands the task to a reviewer agent, which runs typecheck and tests in a Docker sandbox.
  • Ambient assistants. An always-on agent reads its own goals file every fifteen minutes, decides whether something needs doing, and either acts or backs off.
  • Embedding. Drop the runtime into your own Node service and expose agents over your existing API.

What this isn't

It isn't a hosted chat product. It isn't an MCP server. It isn't an opinionated framework for building a single LangChain-style pipeline. If the thing you want is a chat box, ChatGPT is better at chat than TAI is. TAI shines when the agent needs to do something while you're not watching.

Status

This is v0.1. The runtime, tools, agents, channels, memory, workflows, cron, HTTP API, and the eight extension registries (tool, channel, provider, embedding, task backend, step executor, trigger, skill) ship and are tested. The declarative plugins: [...] loader and built-in HTTP auth are v0.2 work. Pages marked "v0.2" describe finalized interfaces with the loader still landing.

Where to go next

If you have ten minutes: Quick start walks the standup scenario above end-to-end.

If you have an hour and want the full picture: read Architecture, then Agents, then Tools. The rest of the Concepts section can be read in any order.

If you're picking a package to install: Packages overview maps the five @tailored-ai/* packages and when to install which.

If you're extending: Custom tools is the YAML path; Extending in code is the TypeScript path; Plugins is the npm-distribution path (live today via registries; declarative loader v0.2).