aweb

aweb is open communication infrastructure for AI agents: stable identities, durable messages, and wake-up events across sessions, runtimes, and machines. Letting agents talk to each other is one step towards teams of them doing useful work alongside people. It also puts into practice an idea behind much of my writing: an organization’s identities and working relationships should outlast its choice of models and tools.

Documentation · Source code

Tools for AI and automation

These libraries grew out of the automation I built for my work at Quantica and GreaterSkies.

  • llmring: a lockfile and versioned registry for LLM models, so code binds to aliases instead of model IDs. I use it as infrastructure for model integration.
  • llmemory: document memory and retrieval for working with organizational knowledge, developed as part of my AI automation tooling.
  • pgdbm: an async PostgreSQL library with migrations, pooling, testing, and monitoring, built to support the database layer of my applications.

Smaller tools

  • magnitude: a Python library for working with numbers and units.
  • dated-money: date-aware currency operations, built for GreaterSkies' backend and accounting across currencies.

Retired and superseded

  • llmbridge: superseded by llmring.
  • ox-leanpub: exports org-mode files to Leanpub book format; no longer maintained.
  • org-jekyll: exports org-mode files to Jekyll; no longer maintained.

Other work

GreaterSkies is the custom star map business I built and run. My earlier engineering work includes 49 patents in 37 families, mostly in printing and colour science.