meteora pro lab

Meteora Pro Lab is a research group building open-source infrastructure for AI-augmented development.

We are not a startup. We run experiments, publish results as open source, and spawn commercial projects when experiments mature. Distributed team.

Principles

40% self-development — learning by building
30% technology advancement — open source, research, publishing
30% commercial work — to fund the other 70%

Money is a resource, not the goal. Time is what we actually spend.

What we build

devboy-tools

Apache 2.0, Rust

Tool bundle for AI coding agents. Unified CLI and MCP server for GitLab, GitHub, Jira, ClickUp, Confluence, Slack. OS keychain for credentials. TOON pipeline for token-efficient output.

51,077 production MCP calls since December 2025, 17 organizations

github.com/meteora-pro/devboy-tools

lokb

MIT, Rust

Local Offline Knowledge Base. Offline-first personal knowledge base. 15 data formats. Hybrid search (BM25 + FM-index + vector). Knowledge graph. MCP server for Claude Desktop.

github.com/meteora-pro/lokb

devboy-agent-usage

Rust

Analytics tool for Claude Code sessions. Cost, time, tasks, focus analysis from JSONL logs.

github.com/meteora-pro/devboy-agent-usage

Research direction

Memory Fact Language (MFL) — a formal language for AI agent memory. Treats facts as compiler IR: structured entities, typed links, executable rules. Dream cycle optimization passes (deduplication, pattern recognition, temporal compaction) keep knowledge compact and consistent.

Builds on SimpleMem, A-MEM, Graphiti/Zep.

Read ADR-009

Commercial projects

We have several commercial startups in development. Not ready to announce yet — check back later.