Announcing Autobot - AI agent framework in Crystal

1

Hey everyone!

Crystal deserves its own AI agent framework — and it’s a perfect fit. The language’s strengths shine here: efficiency, tiny footprint and real concurrency make it ideal for running autonomous agents.

So I built GitHub - crystal-autobot/autobot: Ultra-efficient personal AI assistant powered by Crystal .

The numbers: 2MB binary, ~5MB RAM, <10ms startup, zero runtime deps. Run dozens of isolated bots on a single machine — each with its own personality, workspace, and config.

Features

  • Multi-provider LLM
  • Chat channels
  • Kernel-enforced sandbox
  • Memory & sessions
  • Cron scheduler
  • Extensible

What’s next

There’s a big roadmap ahead — MCP support, voice support, multi-agent orchestration, and more. If you’re interested in helping build the Crystal AI ecosystem, feel free to reach out. Contributions, ideas, and feedback are all welcome.

7 Likes
2

Autobot v0.3.0 is out!

A lot has happened since first release. Here’s a quick rundown of everything new across v0.1.x, v0.2.0, and v0.3.0:

New features

  • Live documentation: Autobot
  • MCP support – connect external tools via the Model Context Protocol, with a tools allow list for fine-grained control
  • AWS Bedrock provider – use Bedrock as an LLM provider alongside existing options
  • Image support – the agent can now process and respond to images
  • Voice support – voice input/output capabilities for the agent
  • Interactive setup – guided setup wizard to get started faster
  • Cron jobs – scheduled tasks with conditional delivery, delayed execution, and update support

Improvements

  • Better sandboxing and path resolution

  • Improved memory consolidation

  • HTTPS fetching with proper SNI handling

  • Slack threading and allow list fixes

  • Better tool logging and error parsing from providers

  • Custom slash command descriptions

  • Documentation site with search

2 Likes