There is no Batman without Alfred

Alfred is a general-purpose AI agent — a co-conspirator, not a butler. He reasons, acts, remembers, and can extend his own capabilities. Talk to him via Telegram or the web UI. Give him a task; he figures out how to do it.

The Why

Inspired by ClawdBot / OpenClaw, Alfred is the outcome of a few sleepless weeks spent understanding how an agent works and how to build one. I originally tried to fork the OG repo and refactor the codebase into what I wanted. After two weeks of failing because I had very little idea what I was doing, I decided to build Alfred from the ground up.

I took a lot of inspiration from Shantanu Goel’s bot and eventually landed on Alfred — built in TypeScript with a very simple ReAct loop.

The Architecture

There’s a brain, and then there are the arms, legs, and everything else.

Alfred has one reasoning loop, which ingests a request and reasons about it. Then there’s a collection of tools the brain can access. Tool descriptions are inserted at runtime, and the brain can choose which tool to use.

I was very clear that this agent wasn’t supposed to be bound by deterministic guardrails. There’s no true autonomy without risk, so most of the safety is baked into the personality. That also means it isn’t the safest agent. But we’ve got quite a few great features that help ensure Alfred is sane and useful.

While Alfred has a soul and his own personality, the underlying model powering him can still creep through in his behaviour.

Suffice to say, Alfred is quite test-driveable today. But use at your own risk — the repository link is at the top of this page.

What Alfred Does Today

General-purpose ReAct agent — research, writing, lead generation, ops, file work, shell commands
Multi-provider LLM — Gemini, Anthropic, OpenAI; configurable per deployment
Telegram + Web UI — converse from your phone or browser; live progress updates as he works
Persistent memory — session context, conversation window, workspace artifacts
Self-extending — Alfred can read his own codebase and write new tools mid-session
Tool ecosystem — search (SearXNG), web fetch, file read/write/edit, shell exec, lead extraction, writer agent, RAG memory

Interesting Reading

The docs are actually thought through. I spent a lot of time studying Oxydra and put thought into everything being built here.

I’ve tried to keep it very simple, so you might enjoy reading the AlfredAI documentation.